A Hybrid CNN–GRU Framework for Real-Time IOMT Data Processing Using Distributed Apache Spark and Apache Ignite

A Hybrid CNN–GRU Framework for Real-Time IOMT Data Processing Using Distributed Apache Spark and Apache Ignite | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Hybrid CNN–GRU Framework for Real-Time IOMT Data Processing Using Distributed Apache Spark and Apache Ignite Mohamad Karamkish Zahooki, Erfaneh Noroozi, Mehdi Hosseinzadeh This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7564813/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Today, with the increasing development of technologies, the use of the Internet of Medical Things (IOMT) has created a new revolution in healthcare systems; therefore, it is essential to implement an efficient, scalable method for real-time processing of medical big data for disease monitoring, diagnosis, prevention, and the overall maintenance of patient health. In this paper, we present a hybrid deep learning model, "CNN-GRU," based on gated recurrent units (GRU) and convolutional neural networks (CNN) for processing IOMT data with a distributed processing environment leveraging Apache Spark and Apache Ignite. The proposed architecture uses Spark for parallel and distributed data preprocessing and feature extraction, and Ignite for in-memory computation and low-latency storage. The proposed method's results are evaluated using four parameters: precision, recall, accuracy, and F1 score. Within the suggested model, the average accuracy and recall criterion is 99.01%, the precision criterion is 99.06%, and the F1 score is 99%—the execution time of the model in three different environments. The Spark, Ignite, and Ignite combinations are 18.84 seconds, 18.93 seconds, and 18.18 seconds, respectively. These values obtained from the results of the hybrid architecture indicate an increase in the efficiency of the proposed method compared to previous approaches. This hybrid architecture offers a clear perspective for personalized healthcare systems. Hadoop real-time processing medical big data Internet of Medical Things deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 1- Introduction Currently, matters related to health and medicine hold significant importance for individuals. Additionally, today's world faces numerous challenges related to public health, especially with the emergence of threatening infections such as COVID-19 [1]. Surveys carried out during different periods have revealed that diseases, especially infectious and cardiovascular diseases, are the leading cause of human deaths worldwide [2]. People's attention to their physical health is increasing. Healthcare and online health monitoring play a special and decisive role in preserving the lives of humanity [3]. The healthcare industry, like other industries, has undergone enormous changes. Today's modern healthcare system has many advantages. Applying the IOT in healthcare systems can lead to excellent results in this field[4]. Today, with the growth of wireless communication technologies and the expansion of the Internet, IOT technology has flourished, and its impact on healthcare systems can be seen in the improvement of the patient care process and the improvement of disease diagnosis results[5]. The wide range of solutions offered by the IOT enables sustainable healthcare delivery, enhances the standard of healthcare services, and reduces healthcare costs[6]. The emergence of the IOT in medicine dates back to the 1990s. A decade during which the first medical devices connected to the internet, such as heart monitors, appeared. In 2010, the first program of large medical equipment companies, such as IBM, entered the IoT in this field. Four years after the beginning of IOT technology development, it was widely used in healthcare and medical fields in 2014. This year, the public was given access to devices such as smart watches and health bands. Between 2016 and 2018, there was considerable development of IOT technology in medical devices and hospital systems; in 2018, 5G internet technology was used for transmission. Today, the IOT is usually explicitly used in medicine and has made the process of determining, diagnosing, treating, preventing, and controlling the spread of the disease significantly easier[7]. Today, the IOT and artificial intelligence have made significant progress beyond human expectations. The development of machine learning, deep learning techniques, and big data management has significantly enhanced medical tools for the detection and prevention of numerous diseases[8]. With the advent of the IOMT, opportunities have been created to improve human healthcare[9]. IOMT technology offers innovative possibilities for healthcare systems to oversee and manage patient health [10]. In recent times, healthcare systems have adopted wearable sensor devices that are attached to the human body to monitor vital parameters such as body temperature, heart rate, and blood pressure [11]. The IOT in medical systems has a four-stage architecture[12]. The first stage involves deploying networked connected devices. The second step is to convert the data received from sensors and other devices from analog to digital forms. The third step consists of cleaning, normalizing, and standardizing the data[13]. In the fourth step, the final data are analyzed and processed at the required management level and provided to the doctor for effective decision-making. The data presented and processed at this stage can change the process of diagnosis, prevention, and better control of the disease in the healthcare system. Figure 1 presents an overview of the IOMT within healthcare. The implementation of IOMT has resulted in an increased generation of healthcare data [14]. Recent advances in healthcare systems, especially in the field of IOMT, are rapidly growing, and the volume of health data obtained has led to an increasing amount of helpful information in this field[15]. Medical big data refers to an extensive aggregation of health and medical information gathered from multiple sources and locations, such as high-powered equipment, wireless sensor networks, mobile phone applications and sensors, hospitals, clinics, and monitoring tools [16]. The analysis of this medical data is critical. The results of this analysis help us gain the necessary knowledge in the field of patient care and uncover hidden information. These patterns can predict the course of diseases and reduce the death rate of people, and improve the treatment process of patients [17]. Collectively, these factors and their respective strengths contribute to enhancing the quality of healthcare services, lowering expenses, and reducing the likelihood of mistakes caused by human factors [18]. 1-1 Healthcare Big Data (HBD) The concept of big data appears when you are faced with large amounts of data[19]. Big data encompasses the emergence of large-scale, diverse datasets resulting from growing limitations in storage capacity, advancements in data processing capabilities, and enhanced accessibility [20]. Healthcare big data is a vast field that generates data from many sources[21]. Healthcare big data (HBD) can be classified based on the method of its generation as well as the primary origin from which it is produced. First, medical data often comes from medical and public health records. It can also come from wireless sensors or sensor-equipped wearables. Second, public health data encompasses both public health records and other relevant health records. Third, medical images of the inside of the body. Fourth, laboratory and research data[22]. Big data analysis and processing, and knowledge generation in the healthcare system are the result of gathering substantial amounts of information from various sources, such as electronic health records, clinical documentation, omics datasets, and personalized medicine systems[23]. Figure 2 illustrates a category of big data utilized within healthcare. 1-2 Healthcare big data analytics The emergence of big data has challenged data management and accessibility, data security and analysis, and data use in various fields [24]. By analyzing healthcare data and generating knowledge in this field, we can reduce treatment costs, predict epidemics, and improve the quality of human life. Doctors want to know as much as they can about their patients' lifestyles and health. Therefore, they can detect warning signs of a serious illness earlier and better based on the processing of the analyzed vital data of their patient. This data is usually presented in different formats and sizes[25]. In medical and healthcare systems, data analysis encompasses various scientific fields, including bioinformatics, medical imaging, sensors, and medical informatics[26]. Data analytics focuses on the integration of heterogeneous data[27]. The new knowledge discovered through data processing and analysis will have many benefits for patients, doctors, and health policy makers. Figure 3 provides an overview of big data analytics in the healthcare sector. 1-3- The contribution of the paper With the increasing spread of IOMT technology in medicine, there has been a significant increase in networked medical devices that continuously collect substantial volumes of diverse patient data available online and in real time [28]. This voluminous data requires timely analysis to inform effective treatment and decision-making in human healthcare. Big data processing systems such as Hadoop are suitable for batch processing but lack the low-latency processing capabilities required for real-time data streams and may not provide the fault tolerance and scalability needed to manage large volumes of medical big data[29]. This issue can lead to delays or inaccuracies, increasing errors in diagnosing diseases, treatment, and patient monitoring systems, especially in critical care scenarios where timely data analysis can save lives. The motivation of this article is to examine the existing challenges, identify them, and resolve these limitations by developing a hybrid architecture with Apache Spark and Apache Ignite. This architecture combines a deep learning algorithm that addresses Hadoop's weaknesses for large-scale batch processing with real-time data frameworks. This paper introduces a hybrid CNN-GRU framework specifically designed for real-time IOMT data processing within a distributed architecture via Apache Spark and Ignite. The purpose of this hybrid approach is to provide a more efficient, faster, scalable, and reliable solution for managing medical big data in real-time IOMT applications that will ultimately improve disease outcomes and deliver healthcare solutions. In this section, we succinctly outline the contributions made by our research. The principal achievements of our study are as follows: Hybrid Model Design : We present an innovative hybrid framework that integrates deep learning techniques with CNN and GRU in this architecture. With this integration, complex relationships in IOMT data can be analyzed and processed more comprehensively. Distributed Processing Framework : In this method, the integration of batch and real-time processing components is provided to achieve efficient and scalable processing in healthcare applications and healthcare systems via Apache Spark and Apache Ignite. Real-time processing increases the model's capacity to handle large datasets, which is crucial for timely analysis in medical systems and online monitoring. Performance evaluation : We present a detailed performance analysis, which shows that the hybrid model with Apache Spark and Apache Ignite. Moreover, the GRU-CNN demonstrates marked improvements over conventional methods, delivering superior performance in speed, accuracy, and reduced latency. Practical Implications : The proposed model is developed with practical healthcare uses as a primary focus. This approach demonstrates significant promise for enhancing surveillance and supporting informed decision-making in disease diagnosis and prevention by enabling prompt and precise data analysis. Future Research Directions : This work lays the foundation for further exploration into hybrid architectures and distributed systems in healthcare. By identifying challenges and opportunities, we encourage future studies to refine and expand upon our findings, driving innovations in medical data analytics. In summary, our study addresses significant deficiencies in the existing IOMT data processing framework, offering a robust solution that integrates advanced deep learning techniques with Apache Spark and Apache Ignite, featuring a scalable architecture to enhance healthcare delivery and system outcomes. The primary divisions of this article are outlined below: The introduction in Section 1 is followed by background information in Section 2. Section 3 provides an overview of the literature review and highlights key findings pertinent to the subject matter. Section 4 introduces the system model and outlines the proposed method. In Section 5, the evaluation results that were obtained are presented. Finally, Section 6 presents the outcomes derived from the experimental investigations and offers solutions for future research and advancements in this area, along with strategies to further improve the field moving forward. 2- Background This section describes the various tools and technologies essential for processing big data, especially in the healthcare field, where data is gathered from multiple origins, such as medical instruments, patient records, and imaging systems. Advanced tools are crucial for the effective processing, storage, and analysis of the growing complexity and quantity of data generated within the healthcare sector. 2-1- Big data in healthcare systems Today, big data technology is widely used in healthcare systems and many other fields [30]. With the digitization of medical information, the transfer of data from electronic medical records and the analysis of medical data alongside the extraction of knowledge derived from such information have become faster [31]. In the fields of medicine and healthcare, big data analytics encompasses the examination of multiple scientific disciplines, including bioinformatics, medical imaging, and health informatics [32]. Big data analysis techniques reveal new knowledge that provides comprehensive benefits to both patients and physicians [33]. Integration of diverse data extracted from different databases, standardization of protocols, and large amounts of laboratory data are among the challenging issues of healthcare systems [34]. 2-2- Categories of Data in Big Data Framework Traditional processing systems and legacy storage solutions are inadequate for handling or storing big data[35]. Big data is distinguished by the extensive diversity of its data formats. These types of data are usually unstructured, semi-structured, or structured data that are produced from various sources. These data sources include: Traditional enterprise data combines customer or organization data from CRM frameworks [36]. Value-based ERP data, online store transactions, and general ledger data are examples of this type of data. Sensor data and machine - generated data , including contact detail records, blogs, counters, sensors, gear reporting, and exchange frameworks, become information. Information of social environments: Data streams sent by web and digital communication tools, such as instant messaging, Twitter, Facebook, and online networks, are part of this information, are stored in a distributed manner, and are voluminous examples of big data. Stock Trade Data: Stock trade data keeps data about "purchases" and "sales" choices in shared assets; the organization oversees customer offerings. Online buying and selling data from the stock market constitutes an example of this type[37]. Transport Data: The information of transportation systems, including displays, restrictions, separations, access maps, and vehicle routes, can be stored in different diverse databases in a distributed manner or collectively in one place[38]. Search Engine Data: Search engines retrieve a set of different information from different tables and store it in other places in a distributed manner in their big data. These data can be retrieved and used from various sources when needed[39]. 2-3- Big Data Tools and Technologies in Healthcare Systems Findings obtained through the analysis of large-scale data sets within the healthcare industry can play a crucial role in epidemic prevention, enhance the management of diseases, and contribute to lowering healthcare expenditures, among various other advantages[40]. To process this massive amount of data in data centers, cloud service providers typically require an increasing amount of computing power and massive storage infrastructure. Medical data collection requires infrastructure and the assembly of various tools and components to solve big data problems[41]. Apache Hadoop is among the platforms utilized for analyzing large-scale data. Hadoop serves as a platform that enables the processing of large-scale data, allowing organizations to uncover meaningful insights and trends that can enhance business processes [42]. The subsequent section outlines additional tools and platforms commonly utilized in big data processing . Apache YARN: Apache YARN is a distributed platform that operates under an open-source model for managing big data resources and task scheduling within the Hadoop Distributed Processing Framework[43]. Apache Spark: Apache Spark is an open-source platform designed for large-scale data processing in distributed computing environments that performs fast processing and complex analysis[44]. Spark’s ability to store data in memory (RAM) during computations allows it to perform operations much faster than traditional disk-based systems such as Hadoop Map-Reduce. The processing speed of Spark can be significantly increased by reducing the need to write intermediate data to the disk, particularly for repetitive tasks. A primary advantage of Apache Spark lies in its capability to run Spark applications directly on a pre-existing Hadoop cluster. Spark is fundamentally centered around RDDs, which maintain data in memory and provide fault tolerance without relying on data replication[45]. Apache Ignite : Ignite is a distributed computing platform that operates similarly to Spark, but with a key distinction: it is specifically designed to manage and process large amounts of data in real time and boasts high performance for data stream processing[46]. By consolidating the memories of all systems within a network, Ignite provides a global memory for calculations. This allows for faster processing speeds and eliminates the need to consider data storage [47]. For real-time systems with a transactional nature or real-time processing, such as online monitoring systems, patient control and monitoring systems for large medical datasets, stock buying and selling systems, and fraud detection, where transaction-oriented processes are required, Ignite is more suitable[48]. Apache HIVE: Apache Hive is a software initiative developed as part of the Apache Hadoop ecosystem and used for data aggregation, querying, and analysis. Hive has a SQL-like interface that is utilized to retrieve information from databases and file storage systems [49]. Apache Mahout: Apache Mahout is an initiative by the Apache Software Foundation aimed at developing open-source and scalable machine learning algorithm implementations. Originally designed as part of the Apache Hadoop project, Mahout focused on providing algorithms for data clustering[50]. Apache Storm: Apache Storm is excellent at processing real-time data streams, which allows for continuous and unrestricted processing of incoming data[51]. The ability to process millions of tuples (data units) per second per node makes it suitable for applications that need immediate data processing and analysis when it arrives. 2-4- Challenges associated with Big Data in Healthcare Big data refers to a diverse aggregation of data originating from multiple sources and is typically characterized by six distinct attributes [52]. Nevertheless, big data presents a different array of challenges, often encapsulated by the "6 Vs": volume, velocity, variety, variability, veracity, and value [53]. Volume: The high quantity of data that is stored on the systems. Variety: Different types of data. Big data encompasses a wide range of formats, from structured tables and semi-structured XML files to unstructured audio, video, and text. Velocity: In the realm of big data, velocity denotes the rapid pace at which data is produced, gathered, and analyzed [54] . Veracity: Data quality, i.e., the valuable information leading to effective decision-making, is a hallmark of big data. If used properly, big data can solve highly complex problems[55]. Value : within the realm of big data, pertains to the valuable insights and advantages gained through the analysis of extensive datasets. Variability: Big data changes shape and form during processing and the lifecycle, which is also a peculiar characteristic of big data[56]. The six challenges of big data can be leveraged to manage and distribute the vast quantities of information produced within the healthcare sector[57]. The amount of healthcare data that must be gathered and analyzed is significant and constantly growing, and diversity refers to the data that must be collected from various sources[7, 58]. The processing of medical big data, as well as the production of knowledge in this field, should be at a velocity[59]. The trustworthiness of healthcare data is referred to as veracity. Finally, by carefully examining the enormous amounts of data in healthcare, significant information could be discovered. Figure 4 presents the challenges involved in dealing with big data. 2-5- Deep Learning Deep learning represents a significant subfield within machine learning (ML) dedicated to the development of intelligent systems capable of learning from vast amounts of data [60]. Deep learning models are composed of artificial neurons and are arranged in layers. These neurons receive data from the previous layers as input and produce outputs, which are then transferred to the next layer[61]. The initial layer serves to receive input data, and the last layer is responsible for making predictions. Layers that are not visible are known as hidden layers and enable the model to acquire complex data representations[62]. Deep learning models employ a variety of algorithms. This article examines the four leading deep learning algorithms. 2-5- 1 Convolutional Neural Networks ( CNN ) Convolutional neural networks (CNNs) have achieved impressive results. It has emerged as a hallmark in the domain of deep learning for human-computer vision, enabling advancements that were once thought unattainable. These include applications such as facial recognition, autonomous vehicles, intelligent home service systems, and advanced medical diagnostics. Convolutional neural networks, or convolutional networks, consist of several layers. We describe these layers below: Convolutional layer: This layer serves as a core component of CNNs, which are extensively utilized in deep learning, particularly for handling grid-structured data like images and videos. [63]. Modified linear unit (ReLU): A convolutional neural network has a modified linear unit for performing operations on elements. Its output is also a modified coordinate map[64]. Pooling layer: This layer serves as an essential element in CNNs, typically positioned after convolutional layers to decrease the spatial dimensions, specifically the height and width of the input feature maps[65]. Fully connected layer: Fully connected layers (FCs) and convolutional layers are common terms used in CNN due to their overlap. Typically, convolutional layers serve to extract features from the input data, after which FC layers are employed to perform classification . Figure 5 shows a schematic view of CNN [66]. 2-5- 2 Long short-term memory (LSTM) LSTMs represent a specialized form of recurrent neural networks designed to capture and retain information over extended sequences, effectively modeling long-term dependencies [67]. LSTM preserves information over time[68]. These neural networks are useful for predicting time series because they remember the previous input data. LSTM memory networks have a chain-like structure in which four interactive layers each communicate uniquely. In addition to predicting periods. The working steps of LSTM involve, in the first step, forgetting parts unrelated to the previous state[69]. In the next step, they selectively update the cell state values. Finally, in the third step, it is time to output the cell state . The figure below shows how LSTMs work. Figure 6 shows a schematic view of LSTM[70]. 2-5- 3 Gated recurrent unit (GRU) neural network In the previous sections, we investigated simple recurrent neural networks (RNNs) and LSTMs. Simple RNNs suffer from the problem of gradient fading. In contrast, LSTM networks address this issue effectively, but they are computationally heavy and complex. Gated recurrent unit (GRU) neural networks do not encounter the issue of gradient fading and are computationally simple and light[71]. In a GRU, such as an LSTM, only the internal content of the RNN blocks is changed. The GRU is a recurrent network in which the blocks (Units) have been changed as follows[72]. Figure 7 shows a schematic view of the GRU. As is clear from the figure above, in GRU networks, each block, like a simple RNN, receives two inputs, x and h, and performs calculations based on these two outputs. GRU networks can be divided into two sections [73]. The first section is called the "update gate", and the second section of the GRU network is called the reset gate. The task of the "update gate" is to manage past information and determine how much of this information from previous sequences is useful for sending to the next sequence[74]. The "reset gate" controls the extent to which previous information is discarded. Like RNNs and LSTM networks, GRU networks incorporate weights that are determined by compact neural networks within each unit, enabling the model to learn and adjust these weights dynamically [75]. Additionally, as with recurrent networks, the "hidden state (h)" is a vector whose number is determined by the user. As mentioned, these networks are simpler and lighter in terms of calculations than LSTM networks are. Furthermore, they do not have the problem of gradient vanishing, and these networks can be used in long sequences[76]. 2-5- 4 Bidirectional Long Short-Term Memory (BiLSTM) LSTM networks represent a sophisticated type of recurrent neural network, designed to overcome certain limitations typically associated with traditional recurrent neural networks[77]. The cell state of each LSTM unit and its three dedicated gates, which include forgetting, input, and output gates, give the network the ability to oversee the flow of information through it and effectively solve the problem. The vanishing gradients are handled, and long-term features are remembered [78]. BiLSTM is an enhanced deep learning model based on LSTM architecture, capable of analyzing sequential data by traversing it in both forward and backward directions [79]. Its ability to utilize quick round-trip bidirectional information makes it a powerful tool for applications that require complete sequence comprehension. [80] The BiLSTM can better capture patterns and dependencies in data where knowing both the past and the future context is crucial. For example, in natural language processing, understanding a word’s meaning might depend on both the words that come before and those that go after it[81]. The architecture of BiLSTM is as follows : Input Layer : The input sequence is processed by both the forward and backward LSTM. Forward LSTM : This LSTM analyzes the input sequence sequentially, starting at the initial element and proceeding through to the final one [82]. Backward LSTM : This LSTM analyzes the input sequence in reverse order, starting from the last element and moving toward the first [69]. Output Layer : At each time step, the outputs produced by both LSTMs are merged to create the corresponding final prediction. Forward Layer : One LSTM layer processes the sequence in the original order (from time step t=1 to t=k)[83]. Backward Layer : Another LSTM layer processes the sequence in reverse order (from time step t=k to t=1) Combining forward and backward states: At every time step, the hidden states generated by both LSTM layers are merged following the completion of the forward and backward traversals [84]. BiLSTMs are capable of capturing contextual information from both preceding and succeeding elements within a sequence, which enhances their effectiveness in tasks that require comprehensive sequence understanding. As a result, they often surpass traditional LSTMs in applications like machine translation, text generation, and speech recognition, owing to their ability to utilize the complete context of the input sequence[85]. Tasks like named entity recognition (NER), part-of-speech tagging, and sentiment analysis gain advantages from utilizing bidirectional context, as interpreting a word’s meaning frequently relies on the words that come before and after it[86]. Figure 8 shows a schematic view of BiLSTM. 3- Related Work This section examines existing research on real-time processing of medical big data in the context of the IOMT, emphasizing different strategies along with their respective strengths and limitations. The review provides an overview of the current progress in this field and discusses the development of new methodologies. Table 1 summarizes the benefits and drawbacks associated with each approach. In [87] , a new approach and a framework for handling and examining large-scale medical datasets were introduced, exploring the application of Apache Hadoop and Apache Ignite for data management and machine learning within big medical data analysis, as well as assessing the role of artificial intelligence in interpreting analysis outcomes and the associated advantages. The disadvantages of these methods have been discussed. It also addresses business platforms that utilize data results and medical data analytics in various industries, including pharmaceuticals, as well as the challenges of large-scale medical data analysis. The advantages of this research encompass the comprehensive and complete presentation of new methods of analysis and management of big data, along with the advantages and disadvantages mentioned. One of the disadvantages of this study is the lack of detailed analysis of the above methods in data sample processing, and the lack of comparison of the above techniques with each other. [88] provides an overview of the improvement and development of healthcare in China, as well as the risks associated with the privacy and security of patients' medical information. The analysis of significant medical data and the development challenges of the healthcare sector in China has been completed by them. In this article, the risk factors for the protection and confidentiality of healthcare information within big data environments are discussed, along with protective measures tailored to these risks in cloud service environments. Its advantages include the analysis and presentation of exceptional medical data security results in the mobile application section. However, its disadvantages include the lack of a comprehensive algorithm in the data security section, as well as the absence of data encryption and preservation, and the need for improvement methods. It pointed out. In [19], EMR datasets originating from various medical devices were integrated into MongoDB using Hadoop framework processing methods and subsequently analyzed to enhance the efficiency of patient record management. The Map-Reduce framework is also evaluated in this article. The test data in this study were tested on two different Hadoop and MongoDB clusters. Its advantages include the real-time analysis of big data in all three experimental environments and the effective diagnosis and treatment of the disease in the early stages, and its disadvantages include the lack of accurate and transparent analysis and comparisons of the results under different conditions between these platforms. In [89] , an alternative approach to multistage analysis was presented. Image processing within big data environments is facilitated through the Apache Hadoop ecosystem and delivered as a cloud-based service. This study employs a concurrent pipeline execution framework, complemented by a semi-automated system for real-time monitoring, which can identify outlying points without the need to fully execute the first stage, resulting in increased speed and performance of multistage analysis. It achieves guaranteed results and optimal quality, which is one of its advantages over previous methods. Projects include enhancing the speed of the medical image processing pipeline and enabling the simultaneous execution of procedures and diagnostics for early detection of abnormalities. For this purpose, by collecting the results in the first stage, the analysis in the second stage can be implemented based on all the results collected up to that point. One of the disadvantages of this research is its dependence on the results. If the results of the first stage have errors, the execution time will be extended for reanalysis. Among the other disadvantages of the method used in this article, the simulation presented here, which is in the form of text, requires the retransformation of data extracted from big data and spends more time on data transformation. In [90] , a parallel algorithm for microwave image reconstruction was developed utilizing Apache Spark within high-performance computing environments, including both traditional HPC systems and Google Cloud Infrastructure. In this method, the input data is input into a flexible distributed dataset and then divided into several smaller nodes via a cloud-based distributed method. In this study, the parallel imaging algorithm proposed in this article has a 128% improvement in execution speed. However, a disadvantage of the proposed method is the necessity to retrieve information back into memory during each iteration of the image reconstruction process, leading to increased network traffic and slower access to relevant image data. In [91] , a cloud platform was developed to accelerate statistical computations and analysis through big data processing frameworks. One of the advantages of this platform is that it can easily be transformed into dozens of smaller cluster nodes in Spark using cheaper hardware resources and performing processing operations at a better speed. Its disadvantages include the small experimental data sample and the web platform used to evaluate the results. Here, it uses small static data, and this data evaluation platform is unsuitable for processing big data whose data streams are real-time. [92] provides a comprehensive overview of real-time data processing, introducing fundamental concepts and terminology. It explores commonly used technologies for real-time data analysis, reviews prevalent NoSQL storage options suitable for concurrent data environments, and highlights major application areas where real-time data processing is utilized. The main aim of this study is to conduct a comparative analysis of real-time data processing and NOSQL technologies . Among its advantages, we can point out a detailed case study and full descriptions of the methods, and its disadvantages include the lack of a thorough analysis of real data in the test environment for each of the platforms mentioned in this article. In [93] , a detailed description of two advanced and popular streaming frameworks, Apache Ignite and Hazelcast, was introduced. This article also compares Apache Ignite and Hazelcast to provide Java developers, enterprises, and business owners with the knowledge to accelerate their application development. Among its advantages is the introduction of a novel technique that utilizes in-memory processing of real-time data streams. However, a notable limitation of this method is the absence of result analysis within a real-time testing environment. In [94] , A novel and efficient parallel implementation approach utilizing Spark and Radoop has been introduced to address and reduce the substantial computational demands. The primary objective of this research is to markedly decrease the algorithm’s processing time relative to the serial SVM network, while also maintaining high classification performance. This parallel optimization strategy for SVM hyperparameters was executed using Apache Spark in conjunction with the Radoop platform, leveraging distributed data storage through HDFS. One of the advantages of this article is the method of using Spark, and one of its disadvantages is the failure to state the implementation of the combined process of using radioLoop and the inability to state the implementation of the algorithms mentioned in the article. In [95] , A thorough evaluation of big data computing was performed using Apache Ignite and Apache Spark. The comparison focused on four key aspects: the testing environment, available features, supported operations and queries, as well as performance metrics and execution duration. The advantages of this project include the different examinations of each platform under various conditions, and its disadvantages. The combined processing method is not provided. [96] presents the design and implementation of a scalable analytics platform for medical data utilizing Hadoop and Apache Spark. The system addresses challenges associated with data acquisition, distributed storage, data cleansing, and advanced secondary analysis. Notable benefits stem from the integration of machine learning techniques, including classification, clustering, and collaborative filtering algorithms, in conjunction with Apache Hadoop, Apache Spark, and Apache Mahout. Its disadvantages include the lack of transparency in implementing algorithm codes and in implementing and analyzing the data in this article. In [97] , a novel architecture for real-time health prediction via big data was introduced. The system uses a published ML model that is passed to Spark via Kafka topics. This approach leverages streaming data to initiate distributed resources that represent various disorders. By combining Spark streaming and Kafka streaming, the suggested solution provides efficient processing and monitoring of healthcare big data. The proposed model offers multiple benefits, notably the integration of a hybrid approach that merges decision tree algorithms with Spark technology. The introduced component is evaluated based on its quality and subsequently assigned to an appropriate class set. The machine learning framework employed, specifically the decision tree (DT) model, organizes the data into distinct categories. Classification and prediction tasks are executed using the DT model, which supports both binary and multiclass classification through Spark MLlib. One of the disadvantages of this work is that in this integrated approach, there is no comparative analysis between the results obtained from different algorithms. [98] provides an extensive examination of the field, addressing the tools, methods, and associated challenges of applying advanced statistical techniques to data processing, analysis, and outcome prediction. The process uses machine learning algorithms to analyze large datasets, including EHR, medical imaging, and real-time data. The advantages of this research are that it examines all the problems and provides appropriate solutions for the analysis and prediction of diseases. One limitation of the approach presented in this article is the absence of a comparison with earlier methodologies. Additionally, the implementation algorithms associated with the proposed method are not described in detail. In [99] , the design of a biomedical optics system is presented, in which image formation and image analysis are modeled with deep learning (DL). This review summarizes the current flow review of DL work in some of the most active areas of the field, including the recovery of optical properties, fluorescence imaging, and tomography. One of the advantages of this approach is the use of DL for DOI. Among these benefits, a reduction in analysis time, increased quantitative reconstruction quality, and a unique ability to learn complex relationships are noted. Among its disadvantages, we can mention the limited time analysis capability. In [100] , a real-time system was utilized to predict health conditions by processing substantial volumes of medical data within a cloud environment. The approach involves extracting relevant features from the collected data and employing the proposed machine learning algorithm via Apache Spark, resulting in a scalable system that leverages medical parameters for analysis. The strength of this article lies in its implementation of sophisticated machine learning algorithms for the analysis of time series health data, facilitating real-time processing and minimizing latency. However, a notable limitation is the substantial computational power required for these operations. In [101] , a CNN was introduced to autonomously extract features from images, while a LSTM network was utilized for the final classification stage. Furthermore, the hill climbing algorithm (HCA) was employed to optimize key meta-parameters of both the CNN and LSTM, including the dimensions of the convolutional filters in the CNN and the quantity of units within the LSTM layer, while keeping all other variables unchanged. In [102] , an extensive overview of edge computing architectures tailored for IoT applications is provided, detailing the progression of research, practical implementations, future directions, challenges, and unresolved research questions. The paper highlights the advantages of edge computing over traditional cloud computing across various sectors. This work is positioned as a valuable resource for advancing intelligence at the network edge, thereby influencing the development of next-generation innovative edge technologies. One of the advantages of this method is the reduction in delay and the possibility of immediate data processing. However, there are also disadvantages to the technique used in this paper, such as limited processing power at the edge and the possibility of information loss. In [103] , a data mining and machine learning approach was introduced that employs various classification techniques to enhance the classifier’s accuracy. In addition, the combination of various low-performing classifiers with an iterative ensemble approach is used to create a classifier that is both robust and high-precision. This study aims to assess the impact of ensemble and classification machine learning methods on enhancing decision-making processes in cardiovascular healthcare. Through the application of these techniques, the research seeks to connect data-driven insights with clinical implementation, promoting a more proactive, accurate, and patient-focused strategy for managing cardiovascular diseases. The benefits of this methodology encompass greater robustness and higher predictive accuracy, while its drawbacks involve increased computational demands and added complexity in model management. In the review of the papers mentioned above, Researchers have explored a variety of approaches to managing IOMT data, including machine learning algorithms, deep learning methods, distributed data processing platforms such as Hadoop and Spark, as well as various data fusion techniques. Table 1 provides a summary of key advances in this area, highlighting their strengths and limitations. Table 1. Related Work: Processing of Medical Big Data Ref Year Technology Algorithm Advantage Disadvantage [87] 2019 Apache Hadoop and Apache Ignite - The advantages of this article lie in its comprehensive and complete presentation of new methods of analysis and management of big data. Its disadvantages include the lack of detailed analysis of the above methods in data sample processing and the failure to compare these methods. [88] 2020 cloud - The advantages include the analysis and presentation of exceptional medical data security results in the mobile application section. Disadvantages include the lack of a comprehensive algorithm in the data security section, as well as the absence of data encryption, preservation, and improvement methods. [19] 2017 MongoDB, Hadoop, MapReduce The advantages include the real-time analysis of big data in all three experimental environments and the effective diagnosis and treatment of the disease in the early stages. The disadvantages include the lack of accurate and transparent analysis and comparison of results under different conditions between the mentioned platforms. [89] 2018 Cloud, Hadoop - Compared to previous studies and initiatives aimed at accelerating the processing of medical images, this method has the advantage of simultaneously executing diagnostic procedures and tasks to facilitate early detection of abnormalities. Also, to achieve this goal, the results are collected at the initial analysis stage. Among its other disadvantages, we can point to its presentation simulator, which is in the form of text, which requires the retransformation of data extracted from big data. [90] 2020 Apache Spark, Hadoop, Google Cloud - The article’s primary strength is its focus on distributed and parallel computing frameworks as ideal approaches for managing large volumes of unstructured and semi-structured data. It underscores the capabilities of platforms such as Apache Hadoop and Apache Spark in efficiently processing these substantial datasets. The disadvantages of this proposed algorithm are the need to retrieve data into memory during each cycle to reconstruct the resulting image. [91] 2015 MongoDB, Apache Spark The advantages of this platform are that it can easily scale to dozens of Spark cluster nodes using cheaper IT hardware resources. Among its disadvantages, we can point out its test data sample and the web form used to evaluate the results, which uses little static data, which is unsuitable for real-time voluminous data processing. [92] 2022 NoSQL, Apache Spark, Apache Hadoop, Kafka, Flink - Its advantages are that we can point out a detailed case study and a complete description of the methods. The disadvantages include the lack of detailed analysis of real data in the test environment for each of the platforms mentioned in this article. [104] 2022 Apache Ignite, Hazelcast - Its advantages include presenting a new method based on real-time data flow processing in memory. The disadvantages of this article are the lack of analysis of the results in a real test environment. [94] 2020 Spark, Radoop SVM, PGOSVM algorithm The advantages of this article are the method of using Spark and parallel optimization techniques. A disadvantage is the failure to state the implementation of the combined method of using Radoop and the inability to state the implementation of the algorithms mentioned. [95] 2019 Apache Ignite, Apache Spark - The advantages of this project can be the examination of each of the platforms in a different way under different conditions. The disadvantages are that they do not offer a combined processing method. [96] 2018 Apache Hadoop, Apache Spark, and Apache Mahout map reduce Its advantages include the use of combined machine learning and classification and clustering, and collaborative filtering algorithms with Apache Hadoop, Apache Spark, and Apache Mahout. Its disadvantages include the lack of transparency in how to implement algorithm codes and how to implement and analyze data in the article cited. [97] 2024 Hadoop Map-Reduce with Spark decision tree The proposed model has several advantages, including the use of a hybrid method that combines decision tree algorithms with Spark. This new element is classified by considering its quality and grouping it into one of the special sets of classes. Its disadvantages include not comparing the results of different algorithms in this combined method. [98] 2024 Apache Spark, Hadoop , Apache Hive - The advantages of this article are that it provides a review of all the problems and provides appropriate solutions in the analysis and prediction of diseases. Its disadvantages can be pointed out as it is not compared with the previous methods, and the lack of implementation algorithms in each presented method. [99] 2022 CNN, KNN, SVM Deep learning, ML The Advantages of this are the use of DL for DOI and High accuracy in image classification, effective feature extraction. Its disadvantages include the Limited temporal analysis capabilities. [100] 2022 Apache Spark, ML - The strength of this article is its capacity to create sophisticated machine learning models for analyzing time-series health data, thereby minimizing latency and facilitating real-time data processing. Among its disadvantages, it requires significant computational resources. [101] 2023 Hybrid CNN-LSTM Deep learning, The advantages of this article are that it combines spatial and temporal features and improves performance over single models. Among its disadvantages are increased model complexity and longer training times. [102] 2023 Edge computing for real-time analytics - The advantages of this article are that it reduces latency and enables immediate data processing. Among its disadvantages are limited processing power at the edge, potential data loss [103] 2024 Ensemble methods for decision-making decision tree The advantages of this article are enhanced robustness and improved prediction accuracy. The disadvantages of it include an increase in computational overhead and complexity in managing models. The summary of our findings in the review of the mentioned related works includes the following: Deep Learning Techniques : Different deep learning techniques have been utilized to tackle various challenges associated with IOMT data. CNNs have a reputation for processing imaging data, but RNNs, such as GRU, are more skilled at handling time series data. The use of hybrid models that combine both has shown promise in improving overall performance. Distributed Architectures : Efficient handling of large-scale IOT data has been made possible through the use of distributed data processing frameworks, including Apache Spark. These frameworks can facilitate real-time analytics, but they may require significant computational resources and infrastructure. Edge Computing : Edge computing has been the focus of recent studies to reduce latency in real-time analytics. Processing data closer to the source is a way to obtain immediate insights, but this approach can be limited by the processing power available at the edge. Ensemble methods : A proposal has been made to incorporate ensemble techniques to increase prediction accuracy and robustness. However, these methods add to the complexity and computational demands. The reviewed literature highlights the diverse approaches available for real-time IOMT data processing, each with its own strengths and weaknesses. By integrating the complementary capabilities of CNN and GRU into a distributed architecture, there is an opportunity to enhance these methods. In this paper, to fill this gap, we present a hybrid deep learning model, "CNN-GRU", based on GRU and CNN for processing IOMT data with a distributed processing environment leveraging Apache Spark and Apache Ignite. 4- Proposed Model Currently, the healthcare system is essential in people's lives because of the vital role it plays in monitoring people's health. Processing big data in real time is challenging because of the need for error tolerance and information stability. In this work, we employ a combined method using Apache Spark and Apache Ignite to address the data processing challenges of healthcare systems. A proposed system for data processing and monitoring, in which the data sent by the connected devices is first processed in a distributed database and then stored for real-time analysis, is presented . By combining Apache Ignite and Apache Spark with accurate deep learning algorithms on large-scale medical big data, we aim to achieve a better and faster understanding of diseases, improve disease diagnosis, and enhance patient treatment and prevention. In this study, we utilized an integrated approach involving Apache Spark and Apache Ignite to achieve rapid data access. Additionally, deep learning models including CNN and GRU were employed, as detailed in the following section.. Figure 9 illustrates the structure of the system we have developed. 4-1- Hybrid CNN-GRU Algorithm This algorithm is designed for real-time medical Big Data (IOMT) processing by integrating Convolutional Neural Networks (CNN) for spatial feature extraction with Gated Recursive Units (GRU) for time sequence processing. Apache Spark is utilized for distributed data processing, while Apache Ignite is used for real-time data processing and in-memory storage to accelerate data access in this algorithm. The CNN-GRU hybrid algorithm with Spark Apache and Apache Ignite is shown in Figure 10, which is described in the following step-by-step review. 4-2- Dataset Description In this study, we utilize a dataset about heart disease, which encompasses a range of patient attributes including age, sex, blood pressure, cholesterol levels, and additional factors relevant to cardiac health. The dataset comprises 14 distinct features. Our objective is to employ this dataset to estimate the probability of heart disease occurrence in individuals. Through the application of deep learning techniques to these data, it is feasible to uncover patterns linked to heart disease risk, ultimately aiding in enhanced diagnostic accuracy and preventive strategies. 4-3- Data preprocessing In this section, the data undergoes processing to make it suitable for application in deep learning models. Within this study, preprocessing played a crucial and significant role in guaranteeing that large-scale medical data obtained from IoT devices were clean, consistent, and adequately prepared for utilization in deep learning algorithms. Preprocessing techniques are needed to ensure the compatibility of the algorithms used with medical data that comes in different forms, such as image data, classified features, and unstructured text. This article discusses these techniques in detail. The preprocessing steps are as follows: 4-3-1- Drop faulty data At this phase, data that are incomplete or invalid and could adversely affect the model’s effectiveness are detected and eliminated. Data containing missing or incorrect values should be deleted or corrected to improve the model's prediction accuracy. 4-3-2- Grouping features At this stage, the data features are categorized based on their type to facilitate easier processing in subsequent steps. This categorization allows us to apply appropriate strategies for each data type. 4-3-3- Numerical features Numeric characteristics, including age, blood pressure, cholesterol levels, and heart rate, are numerical. For these features to be processed correctly in the model, scaling is necessary. The min–max scaler technique is used for scaling, and the data are mapped to a range between 0 and 1. 4-3-4- Categorical features Categorical variables such as age, blood pressure, cholesterol levels, and heart rate are transformed into numerical representations using label encoding methods. This preprocessing step facilitates the integration of these features into deep learning models, thereby enhancing the models’ predictive accuracy and overall performance in heart disease detection. 4-4- System Setup In this research, a hybrid architecture was employed to optimize the real-time processing of medical big data in the IOT. The system used for implementation is configured as follows in Table 2. Table 2. Hardware and Software Configurations In this work, four client systems are utilized within a cluster network for real-time data processing. Clients receive their data in a distributed manner and perform processing operations on it. The client systems used for implementation are configured as follows in Table 3. Table 3. Four client hardware and software configurations 4-5- Configuration Models In this study, several deep learning models were configured to diagnose diseases from big medical data collected through IOT devices. The models were carefully selected and tuned to ensure optimal performance in real-time medical data processing. The configuration of these models included setting appropriate meta-parameters, feature selection, and training methods to maximize accuracy and efficiency in disease diagnosis. The details of the models are presented in Table 4. Table 4. Model setting specifications 4-6- Data Splitting In this study, three distinct databases were employed for disease detection, with the dataset divided into 70% for training purposes and 30% for testing. The data from these sources were separated into training and testing subsets to enable an effective assessment of the model’s performance under practical conditions. By allocating 70% of the data for model training and reserving 30% for evaluation, the approach provides sufficient information for the model to learn while maintaining a dedicated portion to test its ability to generalize to new, unseen cases. 4-7- Performance metric Several evaluation criteria can be used to measure and scrutinize the performance of a given classification algorithm. Thus, choosing the appropriate criteria depends on the problem we are dealing with. In some cases, accuracy may be the proper choice, but in other cases, recall or precision may be more appropriate for this purpose. Since we are dealing with medical cases, we can use recall (Sensitivity or True Positive Rate) as a performance metric to choose our classifier. The evaluation metrics of accuracy, recall, precision, and F1 score are used. The performance of the disease detection system is assessed via formulas 1 to 4. 4-8- Building the Hybrid Model This section presents the architecture of a hybrid model grounded in deep learning principles. Fundamental deep learning models are integrated with Apache Spark and Apache Ignite to enhance the efficiency of real-time processing for large-scale medical data within the IOMT framework. This architecture makes use of both spatial and temporal analysis capabilities to ensure efficient large-scale data processing. The proposed hybrid model, which combines the strengths of CNNs and GRUs, can create a powerful architecture that can handle various types of big medical data. CNN layers are utilized for processing and extracting features from data, while GRU layers focus on interpreting sequential information from sensor readings and patient health records. Apache Spark stands out as a leading framework for data processing. Its distributed computing architecture enables efficient and scalable management of extensive medical datasets. Spark facilitates operations including data preprocessing, distributed storage, and the parallel execution of machine learning tasks, such as training models based on CNN and GRU architectures. Spark’s MLlib is utilized for preliminary model training, data segmentation, and the real-time analysis of large-scale medical data streams. Apache Ignite's in-memory computing capabilities help speed up data access and reduce latency in data retrieval and processing. Ignite and Spark are seamlessly integrated to provide low-latency storage and cache medical big data in memory for faster computation. The architecture of the hybrid model is outlined as follows: Input Layer: The input layer receives diverse medical data streams: spatial data from medical imaging systems and temporal data from IoT-enabled wearables and sensors. CNN Module: The spatial data are processed through multiple CNN layers to extract high-dimensional features, identifying critical medical conditions such as tumors or fractures in images. GRU Module: Simultaneously, the sequential time series data from the sensors are passed through the GRU layers for temporal pattern recognition. Distributed Processing with Spark and Ignite: Spark orchestrates the data flow between the CNN and GRU models, parallelizing operations. Ignite caches intermediate results, accelerating real-time processing and reducing I/O bottlenecks. Output Layer: The final layer integrates the results from the CNN and GRU to generate predictions, such as disease detection or patient health monitoring outcomes. The purpose of this architecture is to gather the best features of each model and optimize the spatial and temporal aspects of large medical datasets for real-time processing. Compared with individual models, this combined approach offers superior performance, increased diagnostic accuracy, and timely insight to healthcare professionals. The proposed model is evaluated in Fig. 11. 4-9-Advantages of the Proposed Hybrid Model Real-time Performance: By utilizing Spark for large-scale distributed data processing and ignoring in-memory data storage, the system is capable of processing complex medical datasets with near-real-time performance. Scalability: Horizontal scaling of the model is accomplished by incorporating additional nodes into Spark clusters, allowing for efficient management of growing amounts of medical big data. Accuracy: The combination of a CNN and a GRU for time series data enhances diagnostic accuracy by considering both spatial and temporal patterns. Low Latency: Low-latency access to frequently accessed data is guaranteed by ignoring in-memory capabilities, which enables faster processing and decision-making in critical healthcare applications. By integrating the deep learning strengths of CNN and GRU with the distributed computing capabilities of Apache Spark and Apache Ignite, the proposed hybrid model efficiently manages the real-time processing of large-scale medical data. The efficient and real-time processing of medical big data is crucial for improving patient outcomes in modern healthcare systems, which is why this architecture is well-suited for these patients. 5-Result and discussion 5-1-Result In this section, the performances of the CNN, GRU, LSTM, and BILSTM deep learning models, as well as the proposed model, are reported in Tables 5 to 9. Additionally, general diagrams of the models' efficiency evaluation and execution time are shown. Table 5. Results of the CNN CNN Class Accuracy Precision Recall F1 Score Execution Time of spark (s) Execution Time of Ignite (s) Execution Time of Hybrid (s) 0 99.79 90.2 90.79 90.49 11.01 7.51 9.68 1 90.38 90.97 90.38 90.68 average 90.59 90.58 90.59 90.58 The performance results of the CNN model in disease diagnosis classification of two different classes are shown in Table 5. The accuracy, precision, recall, and F1 score columns are used to evaluate the performance quality of the model. The average accuracy and recall criteria for the whole model are 90.59%, and the precision and F1 score criteria are 90.58%, which are close to 90.59%. Additionally, the execution times of the model in three different environments, Spark, Ignite, and a combination of Ignite-Spark, are 11.01 seconds, 7.51 seconds, and 9.68 seconds, respectively. Table 6. Results of the GRU GRU Class Accuracy Precision Recall F1 Score Execution Time of spark (s) Execution Time of Ignite (s) Execution Time of Hybrid (s) 0 96.05 100 96.05 97.99 24.38 23.13 24.06 1 100 96.3 100 98.11 average 98.03 98.15 98.03 98.05 Table 6 shows the performance results of the GRU model in the classification of two different classes of disease diagnosis. Accuracy, precision, recall, and F1 score columns are used to evaluate the performance quality of the model. The average accuracy and recall criterion for the whole model is 98.03%, the precision criterion is 98.15%, and the F1 score is 98.05%. The model's execution time was evaluated in three different environments: Spark, Ignite, and the combined Ignite-Spark, with times of 24.38, 23.13, and 24.06 seconds, respectively. Table 7. Results of LSTM LSTM Class Accuracy Precision Recall F1 Score Execution Time of spark (s) Execution Time of Ignite (s) Execution Time of Hybrid (s) 0 98.03 98.03 98.03 98.03 99.76 79.72 92.76 1 98.08 98.08 98.08 98.08 average 98.05 98.05 98.05 98.05 Table 7 shows the performance results of the LSTM model in the classification of two different classes of disease diagnosis. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average of all the requirements for the LSTM model is equal to 98.05%. The running time of the model has been checked in three different environments: Spark, Ignite, and combined Ignite-Spark, which are 99.76, 79.72, and 92.76 seconds, respectively. Table 8. Results of the BILSTM BILSTM Class Accuracy Precision Recall F1 Score Execution Time of spark (s) Execution Time of Ignite (s) Execution Time of Hybrid (s) 0 96.05 96.05 96.05 96.05 141.91 148.32 153.5 1 96.15 96.15 96.15 96.15 average 96.1 96.1 96.1 96.1 Table 8 shows the performance results of the BILSTM model in the classification of two different classes of disease diagnosis. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average of all the requirements for the BILSTM model is 96.10%. The running time of the model has been checked in three different environments: Spark, Ignite, and combined Ignite-Spark, which are 141.91, 148.32 seconds, and 153.50 seconds, respectively. Table 9. Results of the proposed model (CNN+GRU) BILSTM Class Accuracy Precision Recall F1 Score Execution Time of spark (s) Execution Time of Ignite (s) Execution Time of Hybrid (s) 0 98.03 100.00 98.03 99.00 18.84 18.93 18.18 1 100.00 98.11 100.00 99.05 average 99.01 99.06 99.01 99.03 Table 9 displays the performance outcomes of the proposed model in the classification of two distinct disease diagnosis classes. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average accuracy and recall criterion for the proposed model are 99.01% and 99.06%, respectively, with a F1 score of 99.03%. The execution time of the model has been checked in three different environments: Spark, Ignite, and the combination of Ignite-Spark, which are 18.84, 18.93 seconds, and 18.18 seconds, respectively. Figure 12 shows an evaluation of the deep learning models with four main criteria. As shown in Figure 11, the CNN and BILSTM methods are less efficient than the other methods. The GRU and LSTM models exhibit similar performances, whereas the CNN-GRU (Proposed Model) outperforms the different models. Figure 12 illustrates the performance of various deep learning models, demonstrating the clear advantage of the proposed approach. The times of the calculations and execution of the models are very different. Figure 13 shows that the CNN and GRU models take less time than the other methods do in different Spark, Ignite, and hybrid environments. Consequently, a hybrid approach integrating both CNN and GRU architectures was proposed, demonstrating greater efficiency relative to alternative techniques. When explicitly compared to the GRU model alone, this combined model exhibits superior efficiency percentages. Compared with the GRU model, the execution time of the proposed hybrid model has been improved by nearly 13%, 12%, and 19%, respectively, in different Spark, Ignite, and Ignite-Spark environments, which shows the high efficiency of computing time and faster execution. Next, the proposed model (CNN-GRU) is compared with other methods in Table 10. It can be seen that the proposed model achieves a higher accuracy compared to the different models. Table 10. Evaluation of the proposed model with other methods. Ref Year Method Accuracy [105] 2020 Deep Neural Network 98.10 [106] 2020 CNN + LASSO 85.70 [107] 2021 CNN 97.00 [108] 2021 DL-based Classifier 94.20 [109] 2022 Bi-LSTM 98.80 [110] 2022 Ensemble classifier 89.00 [111] 2022 XGBoost with RFE 95.60 [112] 2023 RF 97.00 [113] 2023 Hybrid CNN-LSTM 74.20 [114] 2024 Optimized SVM 96.00 [115] 2024 Optimized XGBoost 98.40 Current research Proposed model (CNN-GRU) 99.01 5-2-Discussion In this section, we provide a comparative analysis of the execution times observed across five distinct models using three processing systems (Spark, Ignite, and Hybrid). These models include CNN, GRU, LSTM, BILSTM, and the Proposed Model, which are typically used for data analysis, particularly in deep learning tasks. CNN: In the CNN model, the execution time with Ignite (7.51 seconds) is clearly faster than Spark (11.01 seconds). This time difference is primarily due to Ignite’s in-memory data storage, which enables faster processing for smaller or real-time data processing tasks. However, the execution time using the Hybrid method (9.68 seconds) is quite close to Ignite, indicating that the hybrid system performs almost as efficiently as Ignite, with a slight overhead. This suggests that combining both systems results in optimized performance despite some additional coordination overhead. GRU: For the GRU model, the execution times for Spark (24.38 seconds) and Ignite (23.13 seconds) are pretty close. Here, Ignite remains slightly faster than Spark, likely due to its quicker data retrieval from in-memory storage. The execution time for the Hybrid method (24.06 seconds) is somewhat longer than Ignite, but still comparable. This increase in time is attributed to the coordination overhead when combining the two systems. However, when both systems work together in parallel processing, the benefits often outweigh the marginal increase in time. LSTM: In the LSTM model, which typically requires processing longer and more complex data sequences, ignite (79.72 seconds) is faster than Spark (99.76 seconds). This difference suggests that Ignite is more efficient at handling memory-intensive tasks and real-time data processing. However, the execution time with the Hybrid method (92.76 seconds) is slightly slower than Ignite. This is due to the additional overhead from coordinating and synchronizing tasks between the two systems. Nonetheless, this slight increase in time is still justified by the combined advantages of both systems in handling large-scale and complex data processing tasks. BILSTM: In the BILSTM model, which involves processing complex and lengthy data sequences, Spark (141.91 seconds) is faster than Ignite (148.32 seconds). The time difference between these two systems reflects the increased computational requirements of the model. However, the execution time with the Hybrid method (153.5 seconds) is the highest among the three, reflecting the additional coordination and resource synchronization overhead when combining both systems. Although this results in a slightly slower execution time, the hybrid method's ability to handle both extensive data processing and real-time tasks more flexibly still makes it a valuable option for complex models like BILSTM. Proposed Model: In the Proposed Model, the execution times for both Spark (18.84 seconds) and Ignite (18.93 seconds) are almost identical, showing that both systems perform similarly in this case. However, the execution time for the Hybrid method (18.18 seconds) is slightly faster than both, which indicates that the combination of both systems in this model has been particularly optimized to minimize overhead and improve execution time. The Hybrid approach leverages the advantages of Spark for handling extensive data processing tasks alongside Ignite’s capabilities for real-time, in-memory computations. By integrating these technologies, the hybrid system demonstrates notable efficiency in scenarios demanding both large-scale data handling and rapid computational performance, such as with LSTM and BILSTM models. This method contributes to decreased execution times in specific instances while also enhancing the system’s flexibility and scalability when managing substantial and intricate datasets. The Hybrid method is highly scalable, combining the ability of Spark to handle massive datasets with Ignite’s real-time processing capabilities. As a result, the hybrid system offers significant benefits in situations that demand both extensive data processing capabilities and real-time analytical performance. The flexibility to switch between these two systems according to the task at hand allows for optimal performance across different use cases, ensuring that both data volume and processing speed are addressed efficiently. By using a hybrid approach, the system can better manage resources by utilizing Spark for batch processing and Ignite for in-memory real-time processing. This method leads to more optimized resource utilization, as Spark can handle large-scale computations, while Ignite provides fast access to data stored in memory for real-time tasks. This method guarantees efficient resource allocation, resulting in accelerated execution and enhanced system performance, particularly for complex models. Analysis of execution times and evaluation of the advantages provided by the hybrid strategy indicate that, for the majority of models, the Hybrid approach surpasses the performance of using either Spark or Ignite individually, owing to its capacity to leverage the strengths of both platforms. While in some models like BILSTM, the hybrid method results in slightly higher execution times, the overall performance improvement and increased flexibility in handling complex tasks justify this increase in time. 6- Conclusion In this study, we employ Apache Spark and Apache Ignite in combination with advanced deep learning techniques, recognized for their speed and high efficiency in real-time analysis of large-scale medical datasets. We have presented a hybrid model that integrates CNN and GRU architectures, developed explicitly for managing real-time data generated by IOMT devices in healthcare and patient management systems. Our model efficiently manages the intricate nature of IOMT data by employing a CNN to extract relevant features and a GRU to capture temporal relationships, both of which are essential for processing high-dimensional and sequential data characteristics. The scalability and speed of model processing can be significantly improved by implementing a hybrid distributed architecture via Apache Spark and Ignite. This architecture is designed to enable efficient data capture, real-time analysis, and seamless integration into existing healthcare systems, leading to faster and timelier decision-making by physicians and improved patient outcomes. Our experimental results show that the CNN-GRU hybrid model outperforms previous conventional approaches in terms of speed, accuracy, and processing latency, highlighting its suitability for practical applications. Furthermore, the model exhibits strong adaptability across various IOMT environments, as evidenced by its robustness when working with data of different qualities and formats. The objective of this approach was to minimize the processing time involved in real-time handling of large-scale medical data during data retrieval tasks, while simultaneously enhancing execution speed and overall efficiency within systems designed for disease diagnosis and prediction. By combining the deep learning-based CNN and GRU models, we enhance the real-time processing capabilities and accuracy of medical data analysis in healthcare systems. Compared to previous models, our hybrid approach demonstrated improved performance in terms of processing speed, operational efficiency, and accuracy. These improvements suggest that this method has significant potential to advance outcomes in healthcare systems. Based on these findings, we combined a CNN and GRU into a hybrid CNN-GRU model, which achieved improved results in both time and accuracy. The experimental findings highlight the proposed system’s capability to efficiently handle and forecast substantial volumes of real-time medical data originating from various distributed diseases. Assessment of the developed CNN-GRU model demonstrated outstanding results, achieving an accuracy of 99.01%, a recall of 99.01%, a precision of 99.06%, and an F1 score of 99.03%. The model’s performance was further validated by measuring execution times across three different platforms: Spark, Ignite, and their combination, yielding times of 18.84 seconds, 18.93 seconds, and 19.28 seconds, respectively, thereby confirming its effectiveness and efficiency in diverse computational environments. The objective of this approach was to minimize the time needed for real-time processing of large-scale medical data during data retrieval tasks, while simultaneously enhancing execution speed and overall efficiency in systems for disease diagnosis and prediction. By integrating deep learning architectures such as CNN and GRU, we have strengthened both the real-time processing capacity and the accuracy of medical data analysis within healthcare frameworks. When compared to earlier models, our hybrid technique exhibited notable improvements in processing speed, efficiency, and predictive accuracy, positioning it as a promising strategy for advancing healthcare outcomes. This study represents a substantial contribution to the field of real-time IOMT data processing. Despite the surprising progress shown in this study, future work could be done by incorporating other methods, or this hybrid method can be expanded and developed into a more robust, scalable, and secure solution for real-time medical data processing by combining machine learning algorithms alongside CNN and GRU deep learning models. We can improve the system via advanced machine learning techniques. These include ensemble learning, decision trees, and SVMs. The CNN-GRU model can increase accuracy and adaptability. These advancements will increase disease diagnosis efficiency and accuracy. They also improve healthcare management, patient health, and disease control. Declarations Author Contribution Authors:A. Dr. Erfaneh Noroozi)Corresponding Author(B. Phd student- Mohamad Karamkish Zahooki C. Dr. Mehdi Hosseinzadeh Declaration of Competing Interests The authors confirm there are no conflicts of interest associated with this research. Data availability All datasets examined and employed in this study are available from the authors upon reasonable request. The primary dataset analyzed is openly accessible via Kaggle at the following URL: https://www.kaggle.com/datasets/johnsmith88/heart-disease-dataset. References Kashani, M.H., et al., A systematic review of IoT in healthcare: Applications, techniques, and trends. 2021. 192 : p. 103164. Shrotriya, L., et al., Apache Spark in healthcare: Advancing data-driven innovations and better patient care. 2023. 14 (6). Hashemi, S., et al., Service and energy management in fog computing: A taxonomy approach and future directions. 2024. 12 (1): p. 15-38. Hashemi, S.M., et al., Gwo-sa: Gray wolf optimization algorithm for service activation management in fog computing. 2022. 10 : p. 107846-107863. Plageras, A.P., et al. Efficient large-scale medical data (ehealth big data) analytics in the Internet of Things , in 2017, IEEE 19th Conference on Business Informatics (CBI) . 2017. IEEE. Yadav, B., et al., Internet of Medical Things (IOMT) for Chronic Disease Management. 2024. 13 (6): p. 2278-0181. Hashemi, S.M., et al., A new approach for service activation management in fog computing using Cat Swarm Optimization algorithm. 2024: p. 1-36. ME, B.K., Autonomous Intervention Medical Tools . 2024, Worcester Polytechnic Institute. Ahmed, S.F., et al., Insights into Internet of Medical Things (IoMT): Data fusion, security issues and potential solutions. 2024. 102 : p. 102060. El-Saleh, A.A., et al., The Internet of Medical Things (IoMT): opportunities and challenges. 2024: p. 1-18. Pradhan, B., S. Bhattacharyya, and K.J.J.o.h.e. Pal, IoT‐based applications in healthcare devices. 2021. 2021 (1): p. 6632599. Amiri, Z., et al., The deep learning applications in IoT-based bio-and medical informatics: a systematic literature review. 2024. 36 (11): p. 5757-5797. Milasus, E., Design and Implementation of Distributed System Application for Gathering, Harmonizing, and Distributing Data Acquired from Different Data Sources . 2024, Technische Universität Wien. Arshad, H., et al., Trends and Challenges in harnessing big data intelligence for health care transformation. 2024: p. 220-240. Manogaran, G., et al., A new architecture of Internet of Things and big data ecosystem for secured intelligent healthcare monitoring and alerting system. 2018. 82 : p. 375-387. Aminizadeh, S., et al., Opportunities and challenges of artificial intelligence and distributed systems to improve the quality of healthcare service. 2024. 149 : p. 102779. Titus, A., et al., E-Healthcare Data Management Using Machine Learning and IoT , in IoT and ML for Information Management: A Smart Healthcare Perspective . 2024, Springer. p. 167-199. Nazari, E., M.H. Shahriari, and H.J.F. H.I. Tabesh, Big Data analysis in healthcare: Apache Hadoop, Apache Spark, and Apache Flink. 2019. 8 (1): p. 14. Basco, J.A. and N. Senthilkumar. Real-time analysis of healthcare using big data analytics . In IOP Conference Series: Materials Science and Engineering . 2017. IOP Publishing. Ray, P.P.J.F.C. and I. Journal, A survey of IoT cloud platforms. 2016. 1 (1-2): p. 35-46. Wan, K. and V. Alagar. Characteristics and Classification of Big Data in the Healthcare Sector . In 2016, the 12th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD) . 2016. IEEE. Baloch, L., et al., A review of big data trends and challenges in healthcare. 2023. 14 (6): p. 1320-1333. Hassan, S., et al., Big data and predictive analytics in healthcare in Bangladesh: regulatory challenges. 2021. 7 (6). Ahmed, V., et al., The future of big data in facilities management: opportunities and challenges. 2017. 35 (13/14): p. 725-745. Khatiwada, P., et al., Patient-Generated Health Data (PGHD): Understanding, Requirements, Challenges, and Existing Techniques for Data Security and Privacy. 2024. 14 (3): p. 282. Yang, X., et al., Biomedical big data technologies, applications, and challenges for precision medicine: A review. 2024. 8 (1): p. 2300163. Wu, P.-Y., et al., –Omic and electronic health record big data analytics for precision medicine. 2016. 64 (2): p. 263-273. Pradyumna, G., et al., Empowering healthcare with IoMT: Evolution, machine learning integration, security, and interoperability challenges. 2024. Singh, R.K.J.D.L.P., Developing a big data analytics platform using Apache Hadoop Ecosystem for delivering big data services in libraries. 2024. 40 (2): p. 160-186. Cena, F., et al., Introduction to the special issue on big personal data in interactive intelligent systems. 2017. 7 (2): p. 1-5. Pramanik, P.K.D., et al., Healthcare big data: A comprehensive overview. 2022: p. 119-147. Viceconti, M., et al., Big data, big knowledge: big data for personalized healthcare. 2015. 19 (4): p. 1209-1215. El-Gayar, O. and P. Timsina. Opportunities for business intelligence and big data analytics in evidence-based medicine . In 2014, the 47th Hawaii International Conference on System Sciences . 2014. IEEE. Luo, J., et al., Big data application in biomedical research and health care: a literature review. 2016. 8 : p. BII. S31559. Hosen, M.S., et al., Data-Driven Decision Making: Advanced Database Systems for Business Intelligence. 2024: p. 687–704-687–704. Sun, Y., et al., A Real-time Associative Feature-based CRM and ERP Integration Model for SMEs. 2024: p. 1-19. Ahamad, S., et al., Apollo Hospital's Proposed Use of Big Data Healthcare Analytics , in Big Data Analytics Techniques for Market Intelligence . 2024, IGI Global. p. 312-328. Jevinger, Å., et al., Artificial intelligence for improving public transport: a mapping study. 2024. 16 (1): p. 99-158. Salemi, A. and H. Zamani. Towards a search engine for machines: Unified ranking for multiple retrieval-augmented large language models . In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 2024. Chowdhury, A.T., et al., Application of Big Data in Infectious Disease Surveillance: Contemporary Challenges and Solutions , in Surveillance, Prevention, and Control of Infectious Diseases: An AI Perspective . 2024, Springer. p. 51-71. Hashemi, S.M., et al., Energy‐aware resource management in fog computing for IoT applications: A review, taxonomy, and future directions. 2024. 54 (2): p. 109-148. Theodorakopoulos, L., A. Theodoropoulou, and Y.J.E. Stamatiou, A State-of-the-Art Review in Big Data Management Engineering: Real-Life Case Studies, Challenges, and Future Research Directions. 2024. 5 (3): p. 1266-1297. Yao, Y., et al., New scheduling algorithms for improving performance and resource utilization in Hadoop YARN clusters. 2019. 9 (3): p. 1158-1171. Salloum, S., et al., Big data analytics on Apache Spark. 2016. 1 : p. 145-164. Ponnusamy, S. and P.J.I.A. Gupta, Scalable Data Partitioning Techniques for Distributed Data Processing in Cloud Environments: A Review. 2024. Chennuri, S., K.R.J.I.J.O.C.E. Kotha, and TECHNOLOGY, DATA ENGINEERING FOR REAL-TIME ANALYTICS: TURNING DATA INTO INSTANT INSIGHTS. 2024. 15 (5): p. 752-763. Balakrishnan, A.J.I.J.o. C.T. and Technology, Data Virtualization Architecture Framework Using Multi-Engine Data Platforms for Big Data Analytics and Machine Learning. 2024. 72 (2): p. 82-91. AL-SHARGABI, A.A., Blockchain-IoT Healthcare Applications and Trends: A Review. 2024. Talia, D. and P. Trunfio. Programming Tools for High-Performance Data Analysis . In Proceedings of the 33rd International Symposium on High-Performance Parallel and Distributed Computing . 2024. Al-Jumaili, A.H.A., et al., Parallel power load abnormalities detection using fast density peak clustering with a hybrid canopy-K-means algorithm. (Preprint): p. 1-26. Wu, C.-Y., et al. Machine learning-based real-time task scheduling for Apache Storm . In Sensors and Systems for Space Applications XVII . 2024. SPIE. Al-Ali, A., et al., Role of IoT technologies in big data management systems: A review and Smart Grid case study. 2024: p. 101905. Hussein, A.e.e.A.J.J.o.I.S., Fifty-six big data V’s characteristics and proposed strategies to overcome security and privacy challenges (BD2). 2020. 11 (4): p. 304-328. Lipovac, I., M.B.J.I.J.o. D.M. Babac, Modelling, and Management, Developing a Data Pipeline Solution for Big Data Processing. 2024. 16 (1): p. 1-22. Ridzuan, F. and W.M.N.W.J.P.C.S. Zainon, A Review on Data Quality Dimensions for Big Data. 2024. 234 : p. 341-348. Azzam, M.E.A.Y., et al., How big data features drive financial accounting and firm sustainability in the energy industry. 2024. 22 (1): p. 29-51. Bawaneh, R. and A. Jallad. Big Data in Healthcare: A Review on Applications, Technologies, Benefits and Challenges . in 2024 21st International Multi-Conference on Systems, Signals & Devices (SSD) . 2024. IEEE. Ogundipe, D.O.J.I.M.S.R.J., The impact of big data on healthcare product development: A theoretical and analytical review. 2024. 4 (3): p. 341-360. Guo, C. and J. Chen, Big data analytics in healthcare , in Knowledge technology and systems: Toward establishing knowledge systems science . 2023, Springer. p. 27-70. Paramesha, M., N.L. Rane, J.J.P.U.M.R.J. Rane, Big Data Analytics, Artificial Intelligence, Machine Learning, Internet of Things, and Blockchain for Enhanced Business Intelligence . 2024. 1 (2): p. 110-133. Dang, T., et al., Two-layer Ensemble of Deep Learning Models for Medical Image Segmentation. 2024: p. 1-20. Kaveh, A.J.S.i.c.i., Applications of Artificial neural networks and machine learning in Civil Engineering. 2024. 1168 . Jivisar, H., J. Benson, and A. Kowalski, Automated detection of compressed and encrypted ransomware data using a three-layer convolutional neural network. 2024. Seyrek, E.C., M.J.M.T. Uysal, and Applications, A comparative analysis of various activation functions and optimizers in a convolutional neural network for hyperspectral image classification. 2024. 83 (18): p. 53785-53816. Zhao, X., et al., A review of convolutional neural networks in computer vision. 2024. 57 (4): p. 99. Kishore, J. and S.J.M.I.R. Mukherjee, Impact of autotuned fully connected layers on performance of self-supervised models for image classification. 2024: p. 1-13. Le, X.-H., et al., Deep neural network-based discharge prediction for upstream hydrological stations: a comparative study. 2023. 16 (4): p. 3113-3124. Kavitha, D., T. Adilakshmi, and M.C.J.J.o. E.S. Mohan, Preserving Privacy of IoT Healthcare Data using Differential Privacy and LSTM. 2024. 20 (7s): p. 2483-2492. Tzoumpas, K., et al., A data filling methodology for time series based on CNN and (Bi) LSTM neural networks. 2024. 12 : p. 31443-31460. Wang, W., et al., A Hybrid ARIMA-LSTM Model for Short-Term Vehicle Speed Prediction. 2024. 17 (15): p. 3736. Tucker, A.P., et al., Neural Networks With Gated Recurrent Units Reduce Glucose Forecasting Error Due to Changes in Sensor Location. 2024. 18 (1): p. 124-134. Rongali, A.S., et al. ChaoG-GRU: Healthcare Transaction Fraud Detection using Chaotic Iterative Gazelle-based Gated Recurrent Unit Technique . in 2024 Third International Conference on Intelligent Techniques in Control, Optimization and Signal Processing (INCOS) . 2024. IEEE. Alshehri, O.S., O.M. Alshehri, and H. Samma. Blood Glucose Prediction Using RNN, LSTM, and GRU: A Comparative Study . In 2024, the IEEE International Conference on Advanced Systems and Emergent Technologies (IC_ASET) . 2024. IEEE. Tu, B., et al., Real-time prediction of ROP based on GRU-Informer. 2024. 14 (1): p. 2133. Zulqarnain, M., et al., An efficient two-state GRU based on feature attention mechanism for sentiment analysis. 2024. 83 (1): p. 3085-3110. Di Mauro, M., et al. Evaluating Recurrent Neural Networks for the prediction of Multi-Variate time series VoIP metrics . In 2024, the 22nd Mediterranean Communication and Computer Networking Conference (MedComNet) . 2024. IEEE. Ala, A., et al., Enhancing patient information performance in Internet of Things-based smart healthcare system: Hybrid artificial intelligence and optimization approaches. 2024. 131 : p. 107889. Ghosh, S., A. Ekbal, and P. Bhattacharyya, Natural language processing and sentiment analysis: perspectives from computational intelligence , in Computational Intelligence Applications for Text and Sentiment Data Analysis . 2023, Elsevier. p. 17-47. Vishnupriya, G., S. Anusha, and S.J.T.o., E.T.T. Kayikci, An efficient and secure wearable sensor-based remote healthcare monitoring system using adaptive dilated transformer Bi‐LSTM with gated recurrent unit. 2024. 35 (2): p. e4932. Aburass, S., et al., A hybrid machine learning model for classifying gene mutations in cancer using LSTM, BiLSTM, CNN, GRU, and GloVe. 2024. 6 : p. 200110. Mienye, I.D., T.G. Swart, and G.J.I. Obaido, Recurrent Neural Networks: A Comprehensive Review of Architectures, Variants, and Applications. 2024. 15 (9): p. 517. Wu, Z.J.J.o.C.S. and T. Studies, Deep Learning with Improved Metaheuristic Optimization for Traffic Flow Prediction. 2024. 6 (4): p. 47-53. Jerse, G., A.J.A. Marcucci, and Computing, Deep Learning LSTM-based approaches for 10.7 cm solar radio flux forecasting up to 45 days. 2024. 46 : p. 100786. Li, X. and S.J.M. Liu, Predicting 30-day hospital readmission in medicare patients: Insights from an LSTM deep learning model. 2024: p. 2024.09. 08.24313212. Busst, M.M.A., et al., Ensemble BiLSTM: A Novel Approach for Aspect Extraction From Online Text. 2024. Wankhade, M., et al., CBMAFM: CNN-BiLSTM multi-attention fusion mechanism for sentiment classification. 2024. 83 (17): p. 51755-51786. Dash, S., et al., Big data in healthcare: management, analysis and prospects. 2019. 6 (1): p. 1-25. Lv, Z. and L.J.F.G.C.S. Qiao, Analysis of healthcare big data. 2020. 109 : p. 103-110. Bao, S., et al. Technology enablers for big data, multi-stage analysis in medical image processing . In 2018, the IEEE International Conference on Big Data (Big Data) . 2018. IEEE. Ullah, R. and T.J.A.S. Arslan, PySpark-based optimization of microwave image reconstruction algorithm for head imaging big data on high-performance computing and Google Cloud Platform. 2020. 10 (10): p. 3382. Fan, W., D. Zhao, and S. Wang. A fast statistics and analysis solution for medical service big data . In 2015, the 7th International Conference on Information Technology in Medicine and Education (ITME) . 2015. IEEE. Kekevi, U. and A.A.J.C.S. Aydın, Real-time big data processing and analytics: Concepts, technologies, and domains. 2022. 7 (2): p. 111-123. Maxim, B., K.J.I.J.o. C.S. Tetiana, and N. Security, Hazelcast Vs. Ignite: Opportunities for Java Programmers. 2022. 22 (2): p. 406-412. Ali, A.H. and M.Z.J.K.I.J.o. M.S. Abdullah, A parallel grid optimization of SVM hyperparameter for big data classification using Spark Radoop. 2020. 6 (1): p. 3. Stan, C.-S., et al. Apache Spark and Apache Ignite performance analysis . In 2019, the 22nd international conference on control systems and computer science (CSCS) . 2019. IEEE. Rallapalli, S. and R.J.I.J.o. A.E.R. Gondkar, Apache Spark and Hadoop-Based Big Data Processing System for Clinical Research. 2018. 13 (10): p. 7488-7492. Goyal, S.J.A.S. and E.J.F.A. Research, Big Data and Computer-Human Interaction for Real-Time Illness Diagnosis. 2024. 3 (1): p. 20-31. Al-Quraishi, T., et al., Big Data Predictive Analytics for Personalized Medicine: Perspectives and Challenges. 2024. 2024 : p. 32-38. Smith, J.T., et al., Deep learning in macroscopic diffuse optical imaging. 2022. 27 (2): p. 020901-020901. Ebada, A.I., et al., Applying Apache Spark on streaming big data for health status prediction. 2022. 70 (2): p. 3511-3527. Pradhan, A.K., et al., Optimizing CNN‐LSTM hybrid classifier using HCA for biomedical image classification. 2023. 40 (5): p. e13235. Quy, N.M., et al., Edge computing for real-time Internet of Things applications: Future internet revolution. 2023. 132 (2): p. 1423-1452. Kumar Sharma, N., et al., Classification and Ensemble Machine Learning Techniques to Improve Healthcare Decision Making For Heart Disease. 2024. 16 (1): p. 1-11. Bartkov, M., et al., Hazelcast Vs. Ignite: Opportunities for Java Programmers. 2022. 22 (2): p. 406-412. Ayon, S.I., M.M. Islam, and M.R.J.I.J.o.R. Hossain, Coronary artery heart disease prediction: a comparative study of computational intelligence techniques. 2022. 68 (4): p. 2488-2507. Dutta, A., et al., An efficient convolutional neural network for coronary heart disease prediction. 2020. 159 : p. 113408. Mehmood, A., et al., Prediction of heart disease using deep convolutional neural networks. 2021. 46 (4): p. 3409-3422. Kiran, J., N. Debbarma, and S. Ganjala. Heart Disease Prediction Using Machine Learning . In the International Conference on Frontiers of Intelligent Computing: Theory and Applications . 2022. Springer. Nancy, A.A., et al., IoT-cloud-based smart healthcare monitoring system for heart disease prediction via deep learning. 2022. 11 (15): p. 2292. Guleria, P., et al., XAI framework for cardiovascular disease prediction using classification techniques. 2022. 11 (24): p. 4086. Mienye, I.D. y Y. Sun. Effective feature selection for improved prediction of heart disease . In the Pan-African Artificial Intelligence and Smart Systems Conference . 2021. Springer. Yadav, A.L., K. Soni, and S. Khare. Heart disease prediction using machine learning . In 2023, the 14th International Conference on Computing, Communication and Networking Technologies (ICCCNT) . 2023. IEEE. Hossain, M.M., et al., Cardiovascular disease identification using a hybrid CNN-LSTM model with explainable AI. 2023. 42 : p. 101370. Bizimana, P.C., et al., Automated heart disease prediction using improved explainable learning-based technique. 2024: p. 1-30. Mienye, I.D. and N.J.I. Jere, Optimized ensemble learning approach with explainable AI for improved heart disease prediction. 2024. 15 (7): p. 394. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 29 Oct, 2025 Reviews received at journal 15 Oct, 2025 Reviews received at journal 03 Oct, 2025 Reviewers agreed at journal 02 Oct, 2025 Reviews received at journal 18 Sep, 2025 Reviewers agreed at journal 18 Sep, 2025 Reviewers agreed at journal 16 Sep, 2025 Reviewers invited by journal 16 Sep, 2025 Editor assigned by journal 16 Sep, 2025 Submission checks completed at journal 09 Sep, 2025 First submitted to journal 08 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7564813","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":517080279,"identity":"061a2d8f-4de5-4ffc-82cd-68096a023b5a","order_by":0,"name":"Mohamad Karamkish Zahooki","email":"","orcid":"","institution":"Islamic Azad University","correspondingAuthor":false,"prefix":"","firstName":"Mohamad","middleName":"Karamkish","lastName":"Zahooki","suffix":""},{"id":517080280,"identity":"f9e4913d-c920-4c99-9e1b-c4edf15ff785","order_by":1,"name":"Erfaneh Noroozi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/0lEQVRIiWNgGAWjYFACxgdA4gCMJyHHBmXx4NbCbICixZhkLQyJDYScxS99mPFzQcUdBnP23ocPPvyxSO9jP2PA8KOGQcYch2bJvmRm6RlnnjFY9hw3NpzBI5HbxpNjwNhzjIFH5gB2LQZn+A9I87YdZjC4kcYmzSMB1MKQY8DA28DAI4HDYfZnmJl/8/6DaTGQSGfjf2PA+BePFgMeZjZp3gaYlgSJBDaJHANmfLZInGFms+Y5dpjH4MwxZsMZByQM2ySeFRyWOSaBUwt/DzPzbZ6aw3IGx9sYgSFWJy/fn7zx4ZsaG3tcWmAANeIOAK0noGEUjIJRMApGAT4AAMchSOimXO86AAAAAElFTkSuQmCC","orcid":"","institution":"Islamic Azad University","correspondingAuthor":true,"prefix":"","firstName":"Erfaneh","middleName":"","lastName":"Noroozi","suffix":""},{"id":517080281,"identity":"acc77e50-c31b-4270-9c2b-0887f70716d6","order_by":2,"name":"Mehdi Hosseinzadeh","email":"","orcid":"","institution":"Duy Tan University","correspondingAuthor":false,"prefix":"","firstName":"Mehdi","middleName":"","lastName":"Hosseinzadeh","suffix":""}],"badges":[],"createdAt":"2025-09-08 13:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7564813/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7564813/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92167222,"identity":"fdd5226b-700f-4fcc-b4fa-d2f32a73983d","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2219655,"visible":true,"origin":"","legend":"","description":"","filename":"Mkk06007.docx","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/ee696e44e6d0ca7527a11800.docx"},{"id":92168308,"identity":"970cc525-ba78-4095-878d-2b37c9e9f40b","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5206,"visible":true,"origin":"","legend":"","description":"","filename":"dd21c401e7eb4b3ba528ace1af709794.json","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/00060effc1d7e64432fac5ba.json"},{"id":92167231,"identity":"2f198ab3-74db-46ab-b8b6-3c64a40a240c","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":185214,"visible":true,"origin":"","legend":"","description":"","filename":"dd21c401e7eb4b3ba528ace1af7097941enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/b2c669a5d90a7218f548677a.xml"},{"id":92167230,"identity":"88a6829c-fd00-4e5e-b157-343a809c1296","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":103800,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/7473cab5b7dc7799e7ed6aef.jpeg"},{"id":92167241,"identity":"a19f63d5-c31f-4461-9997-e6f59d1571f4","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":881006,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/dc9ee43be42247edb723a552.jpeg"},{"id":92167237,"identity":"31103d86-f908-48b3-bc95-7535127a43dd","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15439,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/a31e8a2f796caed65de612df.png"},{"id":92168326,"identity":"d201f6d3-4732-4bd1-b0d5-33bb95f22008","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15439,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/09f631849ffd4009f80a50c8.png"},{"id":92168329,"identity":"59ca449c-6ecc-44f7-9f54-32d1293f9caf","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46361,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/9045d73685f54b1ccacbea8a.png"},{"id":92167240,"identity":"82df75d1-6a15-420a-8425-f7474d23adc9","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136529,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/89b7b2255978cabf712248d0.jpeg"},{"id":92168332,"identity":"2f982a73-21d8-4ff0-9be1-bbf9014fc63d","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":913314,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/f8242253fdaade208138096b.jpeg"},{"id":92168699,"identity":"95cd504b-3c4f-4fc0-9117-94832c2bdc30","added_by":"auto","created_at":"2025-09-25 11:32:03","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":151636,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/ae5bf0749a134aca5f67497e.jpeg"},{"id":92167257,"identity":"638f4f9d-c3d1-4b58-9bce-09540bab93e6","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70707,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/ad1b85c99e87a44e37ac2c5b.jpeg"},{"id":92167248,"identity":"a561446f-4a05-4980-be24-a67e3b174087","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123643,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/a3c10e4bca96f94c6e76fca0.jpeg"},{"id":92168331,"identity":"7abf23f8-2a37-41db-9255-39e02b3c3244","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94327,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/b1762a5212b62ca408920792.jpeg"},{"id":92167252,"identity":"b7da863f-67ea-4262-8688-8b241371d0a6","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"jpeg","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":278699,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/edc458e5b50a331e6fc2895e.jpeg"},{"id":92167260,"identity":"06059d2f-fbaf-4706-b409-a07afc436bca","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"jpeg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":355756,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/cf412e69b0df1b12c7129e91.jpeg"},{"id":92167245,"identity":"19901f9c-8a32-48ac-8e68-68fbff71634b","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"jpeg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":457469,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/6dfb317362777b614e05d567.jpeg"},{"id":92167243,"identity":"fd708ddc-ed46-4d65-b178-3efb620f31f1","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":51518,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/c08f0591a416d96937da68de.png"},{"id":92167263,"identity":"c0c300a6-7b7e-40c2-a002-5141b78fe749","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":690132,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/9d28b55a5388450db408e532.png"},{"id":92167246,"identity":"d53f8749-d941-45d6-88d3-9b4e9efdf014","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7481,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/3020f0389cb6a2d7fa48d5c7.png"},{"id":92167251,"identity":"560b3e3d-0059-46d3-97ce-ea0a6e2ae29c","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7481,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/9017d1c59008103fd1bf1030.png"},{"id":92168341,"identity":"19acc6a7-3365-4779-a058-07bd18e119e5","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19608,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/30cbef4392d4b4f75712a1bc.png"},{"id":92167266,"identity":"563f25f7-ae75-4ab4-a9ba-c8b3ef4655b2","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28370,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/abb935b5954e4ea5a0c30ea2.png"},{"id":92168340,"identity":"4a12e82c-d2ba-4a69-9caa-1c1f7df1b641","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":174808,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/13c26610c7fe9cf041d6cf2a.png"},{"id":92168337,"identity":"ef34024d-7ce0-43ce-9e5b-4c18addfd83a","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":73932,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/cd0c7e0f6241154c96ff86f0.png"},{"id":92167247,"identity":"76638316-5705-42a9-8e8f-4999973cd6d8","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19136,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/33da93a9b0e89c6739800d21.png"},{"id":92167270,"identity":"01a61fac-5d4c-4fc3-9174-31be4e93e99d","added_by":"auto","created_at":"2025-09-25 11:16:05","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53853,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/a1e37ff42190679276d899c2.png"},{"id":92168338,"identity":"5e1aa2dd-6376-4df1-a63a-c1eb13b45ec2","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":30044,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/0699fc7d1c90a6dd6964ce2b.png"},{"id":92167268,"identity":"4f6f7ea9-77c4-40d4-aa28-74af26efaf6e","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47921,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/6a4f86296b3b187964879536.png"},{"id":92167249,"identity":"e82e187b-80c5-4954-9e7f-43bd1f8c144e","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87338,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/23b99f9d5733b7ff12ea97e3.png"},{"id":92167259,"identity":"1a6e0ba9-cabe-4438-bf14-200bbfc28972","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153919,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/c8045875a6855facf74bdfc5.png"},{"id":92168336,"identity":"b377f360-6d2d-4272-bd68-69b4685761e7","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"xml","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":183609,"visible":true,"origin":"","legend":"","description":"","filename":"dd21c401e7eb4b3ba528ace1af7097941structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/412725854ff85d5d60fc92b5.xml"},{"id":92168339,"identity":"73d1d1b6-fe7e-4904-977c-25c3c3a6cd18","added_by":"auto","created_at":"2025-09-25 11:24:04","extension":"html","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":199792,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/179f5acabb73115a73c26a74.html"},{"id":92168293,"identity":"a2112301-630a-4fc6-a9da-65fbb31ac9b3","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":712060,"visible":true,"origin":"","legend":"\u003cp\u003eIOMT in healthcare.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/c94c5bd08c5532e7b0bd9b01.png"},{"id":92168274,"identity":"87657e9c-d9fb-4179-a0ef-480fb392285a","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":571659,"visible":true,"origin":"","legend":"\u003cp\u003eCategories of Big Data in Healthcare.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/e4d9ea37e8435d5f527c37ad.png"},{"id":92168321,"identity":"4a4ac1a6-80d8-4ab4-b593-9f16f0d4401a","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":370168,"visible":true,"origin":"","legend":"\u003cp\u003eHealthcare Big Data analytics.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/2684373c1435a2a78068c2f5.png"},{"id":92167225,"identity":"579bc202-0020-4e0c-8da1-aa05b918caeb","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":308644,"visible":true,"origin":"","legend":"\u003cp\u003eThe 6 V’s of big data.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/30eeffd9a17561d41017d0bf.png"},{"id":92167224,"identity":"2f0baedd-9fe9-4da2-b79b-fe9392436c65","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":266585,"visible":true,"origin":"","legend":"\u003cp\u003eConvolutional Neural Networks.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/e4fcf933f336fdfe5b807e88.png"},{"id":92168324,"identity":"36e0263d-e800-420f-92c8-fc2560316816","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":246102,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic view of LSTM\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/1cc97d9afc00ceae58de8fe3.png"},{"id":92167228,"identity":"5e988d3e-db62-4283-87f5-71e0f1c35b15","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":143728,"visible":true,"origin":"","legend":"\u003cp\u003eGated recurrent unit (GRU).\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/e01abd4f5539c9d50a736d69.png"},{"id":92168697,"identity":"f89647d5-c7c7-403e-9c23-fa75588574f5","added_by":"auto","created_at":"2025-09-25 11:32:03","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":207257,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic view of BiLSTM.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/bb7773eb247cab6003ac9dda.png"},{"id":92167233,"identity":"8499ed90-4d7a-4a1f-a3a0-c377a1fdb7b1","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":681283,"visible":true,"origin":"","legend":"\u003cp\u003eProposed Architecture hybrid model\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/920fb8c3bb8132824c2ca9cd.png"},{"id":92168698,"identity":"63cc0f8a-1e1b-490b-ad65-1a1de2afaf15","added_by":"auto","created_at":"2025-09-25 11:32:03","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":409509,"visible":true,"origin":"","legend":"\u003cp\u003eCNN-GRU Pseudo Code\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/24a04eff4f7b49d47a5d83c9.png"},{"id":92168328,"identity":"2f11e6a8-ca2a-49d4-948b-3853d0b7237c","added_by":"auto","created_at":"2025-09-25 11:24:03","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":175427,"visible":true,"origin":"","legend":"\u003cp\u003estructural design of the hybrid model.\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/180a492e320770c6109cc321.png"},{"id":92167255,"identity":"3e21dff7-7584-4a2d-8170-d3052ac6571d","added_by":"auto","created_at":"2025-09-25 11:16:04","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":19501,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of deep learning models\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/bbcbbf287f3b0f2c5a012a8f.png"},{"id":92167239,"identity":"93fb417d-cea0-467c-8277-5c0f5d5183cb","added_by":"auto","created_at":"2025-09-25 11:16:03","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":20991,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluating the execution time of the models\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/ab41b071d3d6a8df153614d0.png"},{"id":92169826,"identity":"c767ecac-4cf3-4bef-a0b3-a775bde41566","added_by":"auto","created_at":"2025-09-25 11:40:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6131617,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7564813/v1/9a9d8d6a-15f4-4e2e-af13-5e3ee367222d.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Hybrid CNN–GRU Framework for Real-Time IOMT Data Processing Using Distributed Apache Spark and Apache Ignite","fulltext":[{"header":"1-\tIntroduction","content":"\u003cp skip=\"true\"\u003eCurrently, matters related to health and medicine hold significant importance for individuals. Additionally, today\u0026apos;s world faces numerous challenges related to public health, especially with the emergence of threatening infections such as COVID-19 [1]. Surveys carried out during different periods have revealed that diseases, especially infectious and cardiovascular diseases, are the leading cause of human deaths worldwide [2]. People\u0026apos;s attention to their physical health is increasing. Healthcare and online health monitoring play a special and decisive role in preserving the lives of humanity [3]. The healthcare industry, like other industries, has undergone enormous changes. Today\u0026apos;s modern healthcare system has many advantages. Applying the IOT in healthcare systems can lead to excellent results in this field[4]. Today, with the growth of wireless communication technologies and the expansion of the Internet, IOT technology has flourished, and its impact on healthcare systems can be seen in the improvement of the patient care process and the improvement of disease diagnosis results[5]. The wide range of solutions offered by the IOT enables sustainable healthcare delivery, enhances the standard of healthcare services, and reduces healthcare costs[6]. The emergence of the IOT in medicine dates back to the 1990s. A decade during which the first medical devices connected to the internet, such as heart monitors, appeared. In 2010, the first program of large medical equipment companies, such as IBM, entered the IoT in this field. Four years after the beginning of IOT technology development, it was widely used in healthcare and medical fields in 2014. This year, the public was given access to devices such as smart\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003ewatches and health bands. Between 2016 and 2018, there was considerable development of IOT technology in medical devices and hospital systems; in 2018, 5G internet technology was used for transmission. Today, the IOT is usually explicitly used in medicine and has made the process of determining, diagnosing, treating, preventing, and controlling the spread of the disease significantly easier[7].\u003c/p\u003e\n\u003cp\u003eToday, the IOT and artificial intelligence have made significant progress beyond human expectations. The development of machine learning, deep learning techniques, and big data management has significantly enhanced medical tools for the detection and prevention of numerous diseases[8]. With the advent of the IOMT, opportunities have been created to improve human healthcare[9]. IOMT technology offers innovative possibilities for healthcare systems to oversee and manage patient health [10]. In recent times, healthcare systems have adopted wearable sensor devices that are attached to the human body to monitor vital parameters such as body temperature, heart rate, and blood pressure [11]. The IOT in medical systems has a four-stage architecture[12]. The first stage involves deploying networked connected devices. The second step is to convert the data received from sensors and other devices from analog to digital forms. The third step consists of cleaning, normalizing, and standardizing the data[13]. In the fourth step, the final data are analyzed and processed at the required management level and provided to the doctor for effective decision-making. The data presented and processed at this stage can change the process of diagnosis, prevention, and better control of the disease in the healthcare system. Figure 1 presents an overview of the \u0026nbsp;IOMT within healthcare.\u003c/p\u003e\n\u003cp\u003eThe implementation of IOMT has resulted in an increased generation of healthcare data [14]. Recent advances in healthcare systems, especially in the field of IOMT, are rapidly growing, and the volume of health data obtained has led to an increasing amount of helpful information in this field[15]. Medical big data refers to an extensive aggregation of health and medical information gathered from multiple sources and locations, such as high-powered equipment, wireless sensor networks, mobile phone applications and sensors, hospitals, clinics, and monitoring tools [16]. The analysis of this medical data is critical. The results of this analysis help us gain the necessary knowledge in the field of patient care and uncover hidden information. These patterns can predict the course of diseases and reduce the death rate of people, and improve the treatment process of patients [17]. Collectively, these factors and their respective strengths contribute to enhancing the quality of healthcare services, lowering expenses, and reducing the likelihood of mistakes caused by human factors [18].\u003c/p\u003e\n\u003cp\u003e1-1 Healthcare Big Data (HBD)\u003c/p\u003e\n\u003cp\u003eThe concept of big data appears when you are faced with large amounts of data[19]. Big data encompasses the emergence of large-scale, diverse datasets resulting from growing limitations in storage capacity, advancements in data processing capabilities, and enhanced accessibility [20]. Healthcare big data is a vast field that generates data from many sources[21].\u0026nbsp;Healthcare big data (HBD) can be classified based on the method of its generation as well as the primary origin from which it is produced. First, medical data often comes from medical and public health records. It can also come from wireless sensors or sensor-equipped wearables. Second, public health data encompasses both public health records and other relevant health records. Third, medical images of the inside of the body. Fourth, laboratory and research data[22]. Big data analysis and processing, and knowledge generation in the healthcare system are the result of gathering substantial amounts of information from various sources, such as electronic health records, clinical documentation, omics datasets, and personalized medicine systems[23]. Figure 2 illustrates a category of big data utilized within healthcare.\u003c/p\u003e\n\u003cp\u003e1-2 Healthcare big data analytics\u003c/p\u003e\n\u003cp\u003eThe emergence of big data has challenged data management and accessibility, data security and analysis, and data use in various fields [24]. By analyzing healthcare data and generating knowledge in this field, we can reduce treatment costs, predict epidemics, and improve the quality of human life. Doctors want to know as much as they can about their patients\u0026apos; lifestyles and health. Therefore, they can detect warning signs of a serious illness earlier and better based on the processing of the analyzed vital data of their patient. This data is usually presented in different formats and sizes[25]. In medical and healthcare systems, data analysis encompasses various scientific fields, including bioinformatics, medical imaging, sensors, and medical informatics[26]. Data analytics focuses on the integration of heterogeneous data[27]. The new knowledge discovered through data processing and analysis will have many benefits for patients, doctors, and health policy makers. Figure 3 provides an overview of big data analytics in the healthcare sector.\u003c/p\u003e\n\u003cp\u003e1-3- The contribution of the paper\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003eWith the increasing spread of IOMT technology in medicine, there has been a significant increase in networked medical devices that continuously collect substantial volumes of diverse patient data available online and in real time [28]. This voluminous data requires timely analysis to inform effective treatment and decision-making in human healthcare. Big data processing systems such as Hadoop are suitable for batch processing but lack the low-latency processing capabilities required for real-time data streams and may not provide the fault tolerance and scalability needed to manage large volumes of medical big data[29]. This issue can lead to delays or inaccuracies, increasing errors in diagnosing diseases, treatment, and patient monitoring systems, especially in critical care scenarios where timely data analysis can save lives. The motivation of this article is to examine the existing challenges, identify them, and resolve these limitations by developing a hybrid architecture with Apache Spark and Apache Ignite. This architecture combines a deep learning algorithm that addresses Hadoop\u0026apos;s weaknesses for large-scale batch processing with real-time data frameworks. This paper introduces a hybrid CNN-GRU framework specifically designed for real-time IOMT data processing within a distributed architecture via Apache Spark and Ignite. The purpose of this hybrid approach is to provide a more efficient, faster, scalable, and reliable solution for managing medical big data in real-time IOMT applications that will ultimately improve disease outcomes and deliver healthcare solutions. In this section, we succinctly outline the contributions made by our research. The principal achievements of our study are as follows:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eHybrid Model Design\u003c/strong\u003e: We present an innovative hybrid framework that integrates deep learning techniques with CNN and GRU in this architecture. With this integration, complex relationships in IOMT data can be analyzed and processed more comprehensively.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDistributed Processing Framework\u003c/strong\u003e: In this method, the integration of batch and real-time processing components is provided to achieve efficient and scalable processing in healthcare applications and healthcare systems via Apache Spark and Apache Ignite. Real-time processing increases the model\u0026apos;s capacity to handle large datasets, which is crucial for timely analysis in medical systems and online monitoring.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003ePerformance evaluation\u003c/strong\u003e: We present a detailed performance analysis, which shows that the hybrid model with Apache Spark and Apache Ignite. Moreover, the GRU-CNN demonstrates marked improvements over conventional methods, delivering superior performance in speed, accuracy, and reduced latency.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003ePractical Implications\u003c/strong\u003e: The proposed model is\u0026nbsp;developed with practical healthcare uses as a primary focus. This approach demonstrates significant promise for enhancing surveillance and supporting informed decision-making in disease diagnosis and prevention by enabling prompt and precise data analysis.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFuture Research Directions\u003c/strong\u003e: This work lays the foundation for further exploration into hybrid architectures and distributed systems in healthcare. By identifying challenges and opportunities, we encourage future studies to refine and expand upon our findings, driving innovations in medical data analytics.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp skip=\"true\"\u003eIn summary, our study addresses significant deficiencies in the existing IOMT data processing framework, offering a robust solution that integrates advanced deep learning techniques with Apache Spark and Apache Ignite, featuring a scalable architecture to enhance healthcare delivery and system outcomes. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe primary divisions of this article are outlined below:\u003c/p\u003e\n\u003cp\u003eThe introduction in Section 1 is followed by background information in Section 2. Section 3 provides an overview of the literature review and highlights key findings pertinent to the subject matter. Section 4 introduces the system model and outlines the proposed method. In Section 5, the evaluation results that were obtained are presented. Finally, Section 6 presents the outcomes derived from the experimental investigations and offers solutions for future research and advancements in this area, along with strategies to further improve the field moving forward.\u003c/p\u003e"},{"header":"2-\tBackground","content":"\u003cp\u003eThis section describes the various tools and technologies essential for processing big data, especially in the healthcare field,\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003ewhere data is gathered from multiple origins, such as medical instruments, patient records, and imaging systems. Advanced tools are crucial for the effective processing, storage, and analysis of the growing complexity and quantity of data generated within the healthcare sector.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2-1- Big data in healthcare systems\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e \u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eToday, big data technology is widely used in healthcare systems and many other fields [30]. With the digitization of medical information, the transfer of data from electronic medical records and the analysis of medical data alongside the extraction of knowledge derived from such information have become faster [31]. In the fields of medicine and healthcare, big data analytics encompasses the examination of multiple scientific disciplines, including bioinformatics, medical imaging, and health informatics [32]. Big data analysis techniques reveal new knowledge that provides comprehensive benefits to both patients and physicians [33]. Integration of diverse data extracted from different databases, standardization of protocols, and large amounts of laboratory data are among the challenging issues of healthcare systems [34].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2-2- Categories of\u003cspan dir=\"RTL\"\u003e \u003c/span\u003eData in\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eBig Data\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;Framework\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTraditional processing systems and legacy storage solutions are inadequate for handling or storing big data[35]. Big data is distinguished by the extensive diversity of its data formats. These types of data are usually unstructured, semi-structured, or structured data that are produced from various sources. These data sources include:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eTraditional enterprise data\u003c/strong\u003e combines customer or organization data from CRM frameworks [36].\u003c/li\u003e\n \u003cli\u003eValue-based ERP data, online store transactions, and general ledger data are examples of this type of data.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSensor data and machine\u003c/strong\u003e\u003cstrong\u003e-\u003c/strong\u003e\u003cstrong\u003egenerated data\u003c/strong\u003e\u003cstrong\u003e,\u003c/strong\u003e including contact detail records, blogs, counters, sensors, gear reporting, and exchange frameworks, become information.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eInformation of social environments:\u003c/strong\u003e Data streams sent by web and digital communication tools, such as instant messaging, Twitter, Facebook, and online networks, are part of this information, are stored in a distributed manner, and are voluminous examples of big data.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eStock Trade Data:\u003c/strong\u003e Stock trade data keeps data about \u0026quot;purchases\u0026quot;\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eand \u0026quot;sales\u0026quot; choices in shared assets; the organization oversees customer offerings. Online buying and selling data from the stock market constitutes an example of this type[37].\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eTransport Data:\u003c/strong\u003e The information of transportation systems, including displays, restrictions, separations, access maps, and vehicle routes, can be stored in different diverse databases in a distributed manner or collectively in one place[38].\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSearch Engine Data:\u003c/strong\u003e Search engines retrieve a set of different information from different tables and store it in other places in a distributed manner in their big data. These data can be retrieved and used from various sources when needed[39].\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003e2-3- Big Data Tools and Technologies in Healthcare Systems\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFindings obtained through the analysis of large-scale data sets within the healthcare industry can play a crucial role in epidemic prevention, enhance the management of diseases, and contribute to lowering healthcare expenditures, among various other advantages[40]. To process this massive amount of data in data centers, cloud service providers typically require an increasing amount of computing power and massive storage infrastructure. Medical data collection requires infrastructure and the assembly of various tools and components to solve big data problems[41]. Apache Hadoop is among the platforms utilized for analyzing large-scale data. Hadoop serves as a platform that enables the processing\u0026nbsp;of large-scale data, allowing organizations to uncover meaningful insights and trends that can enhance business processes\u0026nbsp;[42]. The subsequent section outlines additional tools and platforms commonly utilized in big data processing\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;YARN:\u003c/strong\u003e Apache YARN is a distributed platform that operates under an open-source model for managing big data resources and task scheduling within the Hadoop Distributed Processing Framework[43].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;Spark:\u003c/strong\u003e Apache Spark is an open-source platform designed for large-scale data processing in distributed computing environments that performs fast processing and complex analysis[44]. Spark\u0026rsquo;s ability to store data in memory (RAM) during computations allows it to perform operations much faster than traditional disk-based systems such as Hadoop Map-Reduce. The processing speed of Spark can be significantly increased by reducing the need to write intermediate data to the disk, particularly for repetitive tasks. A primary advantage of Apache Spark lies in its capability to run Spark applications directly on a pre-existing Hadoop cluster. Spark is fundamentally centered around RDDs, which maintain data in memory and provide fault tolerance without relying on data replication[45].\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache Ignite\u003c/strong\u003e\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e:\u003c/span\u003e\u003c/strong\u003e Ignite is a distributed computing platform that operates similarly to Spark, but with a key distinction: it is specifically designed to manage and process large amounts of data in real time and boasts high performance for data stream processing[46]. By consolidating the memories of all systems within a network, Ignite provides a global memory for calculations. This allows for faster processing speeds and eliminates the need to consider data storage [47]. For real-time systems with a transactional nature or real-time processing, such as online monitoring systems, patient control and monitoring systems for large medical datasets, stock buying and selling systems, and fraud detection, where transaction-oriented processes are required, Ignite is more suitable[48].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;HIVE:\u003c/strong\u003e Apache Hive is a software initiative developed as part of the Apache Hadoop ecosystem and used for data aggregation, querying, and analysis. Hive has a SQL-like interface that is utilized to retrieve information from databases and file storage systems [49].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache Mahout:\u0026nbsp;\u003c/strong\u003eApache Mahout is an initiative by the Apache Software Foundation aimed at developing open-source and scalable machine learning algorithm implementations. Originally designed as part of the Apache Hadoop project, Mahout focused on providing algorithms for data clustering[50].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eApache Storm:\u003c/strong\u003e Apache Storm is excellent at processing real-time data streams, which allows for continuous and unrestricted processing of incoming data[51]. The ability to process millions of tuples (data units) per second per node makes it suitable for applications that need immediate data processing and analysis when it arrives.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2-4- Challenges\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eassociated with\u003cspan dir=\"RTL\"\u003e \u003c/span\u003e\u003c/strong\u003e\u003cstrong\u003eBig Data in Healthcare\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBig data refers to a diverse aggregation of data originating from multiple sources and is typically characterized by six distinct attributes [52]. Nevertheless, big data presents a different array of challenges, often encapsulated by the \u0026quot;6 Vs\u0026quot;: volume, velocity, variety, variability, veracity, and value [53].\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eVolume:\u0026nbsp;\u003c/strong\u003eThe high quantity of data that is stored on the systems.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVariety:\u003c/strong\u003e Different types of data. Big data encompasses a wide range of formats, from structured tables and semi-structured XML files to unstructured audio, video, and text.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVelocity:\u003c/strong\u003e In the realm of big data, velocity denotes the rapid pace at which data is produced, gathered, and analyzed\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[54]\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVeracity:\u003c/strong\u003e Data quality, i.e., the valuable information leading to effective decision-making, is a hallmark of big data. If used properly, big data can solve highly complex problems[55].\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e within the realm of big data, pertains to the valuable insights and advantages gained through the analysis of extensive datasets.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVariability:\u003c/strong\u003e Big data changes shape and form during processing and the lifecycle, which is also a\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003epeculiar characteristic of big data[56].\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe six challenges of big data can be leveraged to manage and distribute the vast quantities of information produced within the healthcare sector[57]. The amount of healthcare data that must be gathered and analyzed is significant and constantly growing, and diversity refers to the data that must be collected from various sources[7, 58]. The processing of medical big data, as well as the production of knowledge in this field, should be at a velocity[59]. The trustworthiness of healthcare data is referred to as veracity. Finally, by carefully examining the enormous amounts of data in healthcare, significant information could be discovered. Figure 4 presents the challenges involved in dealing with big data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2-5- Deep Learning\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDeep learning represents a significant subfield within machine learning (ML) dedicated to the development of intelligent systems capable of learning from vast amounts of data [60]. Deep learning models are composed of artificial neurons and are arranged in layers. These neurons receive data from the previous layers as input and produce outputs, which are then transferred to the next layer[61]. The initial layer serves to receive input data, and the last layer is responsible for making predictions. Layers that are not visible are known as hidden layers and enable the model to acquire complex data representations[62]. Deep learning models employ a variety of algorithms. This article examines the four leading deep learning algorithms.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cstrong\u003e2-5- 1\u003c/strong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eConvolutional Neural Networks (\u003c/strong\u003e\u003cstrong\u003eCNN\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConvolutional neural networks (CNNs) have achieved impressive results. It has emerged as a hallmark in the domain of deep learning for human-computer vision, enabling advancements that were once thought unattainable. These include applications such as facial recognition, autonomous vehicles, intelligent home service systems, and advanced medical diagnostics. Convolutional neural networks, or convolutional networks, consist of several layers. We describe these layers below:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConvolutional layer:\u0026nbsp;\u003c/strong\u003eThis layer serves as a core component of \u0026nbsp;CNNs, which are extensively utilized in deep learning, particularly for handling grid-structured data like images and videos.\u003c/p\u003e\n\u003cp\u003e[63].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModified linear unit (ReLU):\u003c/strong\u003e A convolutional neural network has a modified linear unit for performing operations on elements. Its output is also a modified coordinate map[64].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePooling layer:\u003c/strong\u003e This layer serves as an essential element in CNNs, typically positioned after convolutional layers to decrease the spatial dimensions, specifically the height and width of the input feature maps[65].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFully connected layer:\u003c/strong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cstrong\u003eFully connected layers (FCs) and convolutional layers are common terms used in CNN due to their overlap.\u0026nbsp;\u003c/strong\u003eTypically, convolutional layers serve to extract features from the input data, after which FC layers are employed to perform classification\u003cstrong\u003e. Figure 5 shows a schematic view of CNN\u003c/strong\u003e[66].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003cstrong\u003e2-5- 2\u0026nbsp;\u003c/strong\u003eLong\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eshort-term memory\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;(LSTM) \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLSTMs represent a specialized form of recurrent neural networks designed to capture and retain information over extended sequences, effectively modeling long-term dependencies\u0026nbsp;[67]. LSTM preserves information over time[68]. These neural networks are useful for predicting time series because they remember the previous input data. LSTM memory networks have a chain-like structure in which four interactive layers each communicate uniquely. In addition to predicting periods.\u0026nbsp;The working steps of LSTM involve, in the first step, forgetting parts unrelated to the previous state[69]. In the next step, they selectively update the cell state values. Finally, in the third step, it is time to output the cell state\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e The figure below shows how LSTMs work. Figure 6 shows a schematic view of LSTM[70].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003cstrong\u003e2-5- 3\u0026nbsp;\u003c/strong\u003eGated recurrent unit (GRU) neural network\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the previous sections, we investigated simple recurrent neural networks (RNNs) and LSTMs. Simple RNNs suffer from the problem of gradient fading. In contrast, LSTM networks address this issue effectively, but they are computationally heavy and complex. Gated recurrent unit (GRU) neural networks do not encounter the issue of gradient fading and are computationally simple and light[71]. In a GRU, such as an LSTM, only the internal content of the RNN blocks is changed. The GRU is a recurrent network in which the blocks (Units) have been changed as follows[72]. Figure \u003cspan dir=\"RTL\"\u003e7\u003c/span\u003e shows a schematic view of the GRU.\u003c/p\u003e\n\u003cp\u003eAs is clear from the figure above, in GRU networks, each block, like a simple RNN, receives two inputs, x and h, and performs calculations based on these two outputs. GRU networks can be divided into two sections [73]. The first section is called the \u0026quot;update gate\u0026quot;, and the second section of the GRU network is called the reset gate. The task of the \u0026quot;update gate\u0026quot; is to manage past information and determine how much of this information from previous sequences is useful for sending to the next sequence[74]. The \u0026quot;reset gate\u0026quot; controls the extent to which previous information is discarded. Like RNNs and LSTM networks, GRU networks incorporate weights that are determined by compact neural networks within each unit, enabling the model to learn and adjust these weights dynamically [75]. Additionally, as with recurrent networks, the \u0026quot;hidden state (h)\u0026quot; is a vector whose number is determined by the user. As mentioned, these networks are simpler and lighter in terms of calculations than LSTM networks are. Furthermore, they do not have the problem of gradient vanishing, and these networks can be used in long sequences[76].\u003c/p\u003e\n\u003cp\u003e\u003cspan dir=\"\"\u003e\u003cstrong\u003e2-5- 4 \u0026nbsp;\u003c/strong\u003e\u003c/span\u003e\u003cstrong\u003eBidirectional Long Short-Term Memory (BiLSTM)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLSTM networks represent a sophisticated type of recurrent neural network, designed to overcome certain limitations typically associated with traditional recurrent neural networks[77]. The cell state of each LSTM unit and its three dedicated gates, which include forgetting, input, and output gates, give the network the ability to oversee the flow of information through it and effectively solve the problem. The vanishing gradients are handled, and long-term features are remembered [78]. BiLSTM is an enhanced deep learning model based on LSTM architecture, capable of analyzing sequential data by traversing it in both forward and backward directions [79]. Its ability to utilize quick round-trip bidirectional information makes it a powerful tool for applications that require complete sequence comprehension.\u003cspan dir=\"RTL\"\u003e[80]\u003c/span\u003e The BiLSTM can better capture patterns and dependencies in data where knowing both the past and the future context is crucial. For example, in natural language processing, understanding a word\u0026rsquo;s meaning might depend on both the words that come before and those that go after it[81]. The architecture of BiLSTM is as follows\u003cspan dir=\"RTL\"\u003e:\u003c/span\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eInput Layer\u003c/strong\u003e: The input sequence is processed by both the forward and backward LSTM.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eForward LSTM\u003c/strong\u003e: This LSTM analyzes the input sequence sequentially, starting at the initial element and proceeding through to the final one [82].\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBackward LSTM\u003c/strong\u003e: This LSTM analyzes the input sequence in reverse order, starting from the last element and moving toward the first [69].\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eOutput Layer\u003c/strong\u003e: At each time step, the outputs produced by both LSTMs are merged to create the corresponding final prediction.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eForward Layer\u003c/strong\u003e: One LSTM layer processes the sequence in the original order (from time step t=1 to t=k)[83].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBackward Layer\u003c/strong\u003e: Another LSTM layer processes the sequence in reverse order (from time step t=k to t=1)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombining forward and backward states:\u003c/strong\u003e At every time step, the hidden states generated by both LSTM layers are merged following the completion of the forward and backward traversals [84].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBiLSTMs are capable of capturing contextual information from both preceding and succeeding elements within a sequence, which enhances their effectiveness in tasks that require comprehensive sequence understanding. As a result, they often surpass traditional LSTMs in applications like machine translation, text generation, and speech recognition, owing to their ability to utilize the complete context of the input sequence[85]. Tasks like named entity recognition (NER), part-of-speech tagging, and sentiment analysis gain advantages from utilizing bidirectional context, as interpreting a word\u0026rsquo;s meaning frequently relies on the words that come before and after it[86]. Figure 8 shows a schematic view of BiLSTM.\u003c/p\u003e"},{"header":"3- Related Work","content":"\u003cp skip=\"true\"\u003eThis section examines existing research on real-time processing of medical big data in the context of the IOMT, emphasizing different strategies along with their respective strengths and limitations. The review provides an overview of the current progress in this field and discusses the development of new methodologies. Table 1 summarizes the benefits and drawbacks associated with each approach.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[87]\u003c/strong\u003e, a new approach and a framework for handling and examining large-scale medical datasets were introduced, exploring the application of Apache Hadoop and Apache Ignite for data management and machine learning within big medical data analysis, as well as assessing the role of artificial intelligence in interpreting analysis outcomes and the associated advantages. The disadvantages of these methods have been discussed. It also addresses business platforms that utilize data results and medical data analytics in various industries, including pharmaceuticals, as well as the challenges of large-scale medical data analysis. The advantages of this research encompass the comprehensive and complete presentation of new methods of analysis and management of big data, along with the advantages and disadvantages mentioned. One of the disadvantages of this study is the lack of detailed analysis of the above methods in data sample processing, and the lack of comparison of the above techniques with each other.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e[88]\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eprovides an overview of the improvement and development of healthcare in China, as well as the risks associated with the privacy and security of patients\u0026apos; medical information. The analysis of significant medical data and the development challenges of the healthcare sector in China has been completed by them. In this article, the risk factors for the protection and confidentiality of healthcare information within big data environments are discussed, along with protective measures tailored to these risks in cloud service environments. Its advantages include the analysis and presentation of exceptional medical data security results in the mobile application section. However, its disadvantages include the lack of a comprehensive algorithm in the data security section, as well as the absence of data encryption and preservation, and the need for improvement methods. It pointed out.\u0026nbsp;\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/strong\u003e[19],\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eEMR datasets originating from various medical devices were integrated into MongoDB using Hadoop framework processing methods and subsequently analyzed to enhance the efficiency of patient record management. The Map-Reduce framework is also evaluated in this article. The test data in this study were tested on two different Hadoop and MongoDB clusters. Its advantages include the real-time analysis of big data in all three experimental environments and the effective diagnosis and treatment of the disease in the early stages, and its disadvantages include the lack of accurate and transparent analysis and comparisons of the results under different conditions between these platforms.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[89]\u003c/strong\u003e, an alternative approach to multistage analysis was presented. Image processing within big data environments is facilitated through the Apache Hadoop ecosystem and delivered as a cloud-based service. This study employs a concurrent pipeline execution framework, complemented by a semi-automated system for real-time monitoring, which can identify outlying points without the need to fully execute the first stage, resulting in increased speed and performance of multistage analysis. It achieves guaranteed results and optimal quality, which is one of its advantages over previous methods. Projects include enhancing the speed of the medical image processing pipeline and enabling the simultaneous execution of procedures and diagnostics for early detection of abnormalities. For this purpose, by collecting the results in the first stage, the analysis in the second stage can be implemented based on all the results collected up to that point. One of the disadvantages of this research is its dependence on the results. If the results of the first stage have errors, the execution time will be extended for reanalysis. Among the other disadvantages of the method used in this article, the simulation presented here, which is in the form of text, requires the retransformation of data extracted from big data and spends more time on data transformation.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[90]\u003c/strong\u003e, a parallel algorithm for microwave image reconstruction was developed utilizing Apache Spark within high-performance computing environments, including both traditional HPC systems and Google Cloud Infrastructure. In this method, the input data is input into a flexible distributed dataset and then divided into several smaller nodes via a cloud-based distributed method. In this study, the parallel imaging algorithm proposed in this article has a 128% improvement in execution speed.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eHowever, a disadvantage of the proposed method is the necessity to retrieve information back into memory during each iteration of the image reconstruction process, leading to increased network traffic and slower access to relevant image data.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[91]\u003c/strong\u003e\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e,\u003c/span\u003e\u003c/strong\u003e a cloud platform was developed to accelerate statistical computations and analysis through big data processing frameworks. One of the advantages of this platform is that it can easily be transformed into dozens of smaller cluster nodes in Spark using cheaper hardware resources and performing processing operations at a better speed. Its disadvantages include the small experimental data sample and the web platform used to evaluate the results. Here, it uses small static data, and this data evaluation platform is unsuitable for processing big\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003edata whose data streams are real-time.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e[92]\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eprovides a comprehensive overview of real-time data processing, introducing fundamental concepts and terminology. It explores commonly used technologies for real-time data analysis, reviews prevalent NoSQL storage options suitable for concurrent data environments, and highlights major application areas where real-time data processing is utilized. The main aim of this study is to conduct a comparative analysis of real-time data processing and NOSQL technologies\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e Among its advantages, we can point out a detailed case study and full descriptions of the methods, and its disadvantages include the lack of a thorough analysis of real data in the test environment for each of the platforms mentioned in this article.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[93]\u003c/strong\u003e, a detailed description of two advanced and popular streaming frameworks, Apache Ignite and Hazelcast, was introduced. This article also compares Apache Ignite and Hazelcast to provide Java developers, enterprises, and business owners with the knowledge to accelerate their application development. Among its advantages is the introduction of a novel technique that utilizes in-memory processing of real-time data streams. However, a notable limitation of this method is the absence of result analysis within a real-time testing environment.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[94]\u003c/strong\u003e\u003cstrong\u003e,\u003cspan dir=\"RTL\"\u003e \u003c/span\u003e\u003c/strong\u003eA novel and efficient parallel implementation approach utilizing Spark and Radoop has been introduced to address and reduce the substantial computational demands. The primary objective of this research is to markedly decrease the algorithm\u0026rsquo;s processing time relative to the serial SVM network, while also maintaining high classification performance. This parallel optimization strategy for SVM hyperparameters was executed using Apache Spark in conjunction with the Radoop platform, leveraging distributed data storage through HDFS. One of the advantages of this article is the method of using Spark, and one of its disadvantages is the failure to state the implementation of the combined process of using radioLoop and the inability to state the implementation of the algorithms mentioned in the article.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[95]\u003c/strong\u003e\u003cstrong\u003e,\u003c/strong\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp;A thorough evaluation of big data computing was performed using Apache Ignite and Apache Spark. The comparison focused on four key aspects: the testing environment, available features, supported operations and queries, as well as performance metrics and execution duration. The advantages of this project include the different examinations of each platform under various conditions, and its disadvantages. The combined processing method is not provided.\u003c/span\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e[96]\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003epresents the design and implementation of a scalable analytics platform for medical data utilizing Hadoop and Apache Spark. The system addresses challenges associated with data acquisition, distributed storage, data cleansing, and advanced secondary analysis. Notable benefits stem from the integration of machine learning techniques, including classification, clustering, and collaborative filtering algorithms, in conjunction with Apache Hadoop, Apache Spark, and Apache Mahout. Its disadvantages include the lack of transparency in implementing algorithm codes and in implementing and analyzing the data in this article.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[97]\u003c/strong\u003e, a novel architecture for real-time health prediction via big data was introduced. The system uses a published ML model that is passed to Spark via Kafka topics. This approach leverages streaming data to initiate distributed resources that represent various disorders. By combining Spark streaming and Kafka streaming, the suggested solution provides efficient processing and monitoring of healthcare big data. The proposed model offers multiple benefits, notably the integration of a hybrid approach that merges decision tree algorithms with Spark technology. The introduced component is evaluated based on its quality and subsequently assigned to an appropriate class set. The machine learning framework employed, specifically the decision tree (DT) model, organizes the data into distinct categories. Classification and prediction tasks are executed using the DT model, which supports both binary and multiclass classification through Spark MLlib. One of the disadvantages of this work is that in this integrated approach, there is no comparative analysis between the results obtained from different algorithms.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e[98]\u003c/strong\u003e provides an extensive examination of the field, addressing the tools, methods, and associated challenges of applying advanced statistical techniques to data processing, analysis, and outcome prediction. The process uses machine learning algorithms to analyze large datasets, including EHR, medical imaging, and real-time data. The advantages of this research are that it examines all the problems and provides appropriate solutions for the analysis and prediction of diseases. One limitation of the approach presented in this article is the absence of a comparison with earlier methodologies. Additionally, the implementation algorithms associated with the proposed method are not described in detail.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[99]\u003c/strong\u003e, the design of a biomedical optics system is presented, in which image formation and image analysis are modeled with deep learning (DL). This review summarizes the current flow review of DL work in some of the most active areas of the field, including the recovery of optical properties, fluorescence imaging, and tomography. One of the advantages of this approach is the use of DL for DOI. Among these benefits, a reduction in analysis time, increased quantitative reconstruction quality, and a unique ability to learn complex relationships are noted. Among its disadvantages, we can mention the limited time analysis capability.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[100]\u003c/strong\u003e, a real-time system was utilized to predict health conditions by processing substantial volumes of medical data within a cloud environment. The approach involves extracting relevant features from the collected data and employing the proposed machine learning algorithm via Apache Spark, resulting in a scalable system that leverages medical parameters for analysis. The strength of this article lies in its implementation of sophisticated machine learning algorithms for the analysis of time series health data, facilitating real-time processing and minimizing latency. However, a notable limitation is the substantial computational power required for these operations.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[101]\u003c/strong\u003e, a CNN was introduced to autonomously extract features from images, while a LSTM network was utilized for the final classification stage. Furthermore, the hill climbing algorithm (HCA) was employed to optimize key meta-parameters of both the CNN and LSTM, including the dimensions of the convolutional filters in the CNN and the quantity of units within the LSTM layer, while keeping all other variables unchanged.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[102]\u003c/strong\u003e, an extensive overview of edge computing architectures tailored for IoT applications is provided, detailing the progression of research, practical implementations, future directions, challenges, and unresolved research questions. The paper highlights the advantages of edge computing over traditional cloud computing across various sectors. This work is positioned as a valuable resource for advancing intelligence at the network edge, thereby influencing the development of next-generation innovative edge technologies. One of the advantages of this method is the reduction in delay and the possibility of immediate data processing. However, there are also disadvantages to the technique used in this paper, such as limited processing power at the edge and the possibility of information loss.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eIn\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e[103]\u003c/strong\u003e, a data mining and machine learning approach was introduced that employs various classification techniques to enhance the classifier\u0026rsquo;s accuracy. In addition, the combination of various low-performing classifiers with an iterative ensemble approach is used to create a classifier that is both robust and high-precision. This study aims to assess the impact of ensemble and classification machine learning methods on enhancing decision-making processes in cardiovascular healthcare. Through the application of these techniques, the research seeks to connect data-driven insights with clinical implementation, promoting a more proactive, accurate, and patient-focused strategy for managing cardiovascular diseases. The benefits of this methodology encompass greater robustness and higher predictive accuracy, while its drawbacks involve increased computational demands and added complexity in model management.\u003c/p\u003e\n\u003cp\u003eIn the review of the papers mentioned above, Researchers have explored a variety of approaches to managing IOMT data, including machine learning algorithms, deep learning methods, distributed data processing platforms such as Hadoop and Spark, as well as various data fusion techniques. Table 1 provides a summary of key advances in this area, highlighting their strengths and limitations.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e\u0026nbsp; Related Work: Processing of Medical Big Data\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"96%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003eRef\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003eYear\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\u003cstrong\u003eTechnology\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eAlgorithm\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\u003cstrong\u003eAdvantage\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003e\u003cstrong\u003eDisadvantage\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[87]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2019\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Hadoop and\u003cbr\u003eApache Ignite\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; -\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this article lie in its comprehensive and complete presentation of new methods of analysis and management of big data.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eIts disadvantages include the lack of detailed analysis of the above methods in data sample processing and the failure to compare these methods.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[88]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2020\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003ecloud\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;\u003cbr\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;-\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages include the analysis and presentation of exceptional medical data security results in the mobile application section.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eDisadvantages include the lack of a comprehensive algorithm in the data security section, as well as the absence of data encryption, preservation, and improvement methods.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[19]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2017\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eMongoDB, Hadoop, MapReduce\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages include the real-time analysis of big data in all three experimental environments and the effective diagnosis and treatment of the disease in the early stages.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages include the lack of accurate and transparent analysis and comparison of results under different conditions between the mentioned platforms.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[89]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2018\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eCloud, Hadoop\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;\u003cbr\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;-\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eCompared to previous studies and initiatives aimed at accelerating the processing of medical images, this method has the advantage of simultaneously executing diagnostic procedures and tasks to facilitate early detection of abnormalities. Also, to achieve this goal, the results are collected at the initial analysis stage.\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eAmong its other disadvantages, we can point to its presentation simulator, which is in the form of text, which requires the retransformation of data extracted from big data.\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[90]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2020\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Spark, Hadoop, Google Cloud\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;-\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\u0026nbsp;The article\u0026rsquo;s primary strength is its focus on distributed and parallel computing frameworks as ideal approaches for managing large volumes of unstructured and semi-structured data. It underscores the capabilities of platforms such as Apache Hadoop and Apache Spark in efficiently processing these substantial datasets.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages of this proposed algorithm are the need to retrieve data into memory during each cycle to reconstruct the resulting image.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[91]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2015\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eMongoDB, Apache Spark\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this platform are that it can easily scale to dozens of Spark cluster nodes using cheaper IT hardware resources.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eAmong its disadvantages, we can point out its test data sample and the web form used to evaluate the results, which uses little static data, which is unsuitable for real-time voluminous data processing.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[92]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eNoSQL, Apache Spark, Apache Hadoop, Kafka, Flink\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;-\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eIts advantages are that we can point out a detailed case study and a complete description of the methods.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages include the lack of detailed analysis of real data in the test environment for each of the platforms mentioned in this article.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[104]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Ignite, Hazelcast\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;-\u003c/span\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eIts advantages include presenting a new method based on real-time data flow processing in memory.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages of this article are the lack of analysis of the results in a real test environment.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[94]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2020\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eSpark, Radoop\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003eSVM, PGOSVM algorithm\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this article are the method of using Spark and parallel optimization techniques.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eA disadvantage is the failure to state the implementation of the combined method of using Radoop and the inability to state the implementation of the algorithms mentioned.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[95]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2019\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Ignite, Apache Spark\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e-\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this project can be the examination of each of the platforms in a different way under different conditions.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages are that they do not offer a combined processing method.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[96]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2018\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Hadoop, Apache Spark, and Apache Mahout\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003emap reduce\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eIts advantages include the use of combined machine learning and classification and clustering, and collaborative filtering algorithms with Apache Hadoop, Apache Spark, and Apache Mahout.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eIts disadvantages include the lack of transparency in how to implement algorithm codes and how to implement and analyze data in the article cited.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[97]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2024\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eHadoop Map-Reduce with Spark\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003edecision tree\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe proposed model has several advantages, including the use of a hybrid method that combines decision tree algorithms with Spark. This new element is classified by considering its quality and grouping it into one of the special sets of classes.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eIts disadvantages include not comparing the results of different algorithms in this combined method.\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[98]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2024\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Spark, Hadoop\u003cbr\u003e, \u0026nbsp; \u0026nbsp; Apache \u0026nbsp;Hive\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e-\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this article are that it provides a review of all the problems and provides appropriate solutions in the analysis and prediction of diseases.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eIts disadvantages can be pointed out as it is not compared with the previous methods, and the lack of implementation algorithms in each presented method.\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[99]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eCNN, KNN, SVM\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003eDeep learning, ML\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe Advantages of this are the use of DL for DOI and High accuracy in image classification, effective feature extraction.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eIts disadvantages include the Limited temporal analysis capabilities.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[100]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eApache Spark, ML\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e-\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe strength of this article is its capacity to create sophisticated machine learning models for analyzing time-series health data, thereby minimizing latency and facilitating real-time data processing.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eAmong its disadvantages, it requires significant computational resources.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[101]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2023\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eHybrid CNN-LSTM\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003eDeep learning,\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\u003cbr\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003eThe advantages of this article are that it\u0026nbsp;combines spatial and temporal features and improves performance over single models.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e \u003cbr\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\u003cbr\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003eAmong its disadvantages are increased model complexity and longer training times.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e \u003cbr\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\u003cbr\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[102]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2023\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eEdge computing for real-time analytics\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u0026nbsp;-\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003eThe advantages of this article are that it reduces latency and enables immediate data processing.\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eAmong its disadvantages are limited processing power at the edge, potential data loss\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\u003cstrong\u003e[103]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e2024\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003eEnsemble methods for decision-making\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003edecision tree\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003eThe advantages of this article are enhanced robustness and improved prediction accuracy.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e \u003cbr\u003e\u0026nbsp;\u003cbr\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26px;\"\u003eThe disadvantages of it include an increase in computational overhead and complexity in managing models.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe summary of our findings in the review of the mentioned related works includes the following:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eDeep Learning Techniques\u003c/strong\u003e: Different deep learning techniques have been utilized to tackle various challenges associated with IOMT data. CNNs have a reputation for processing imaging data, but RNNs, such as GRU, are more skilled at handling time series data. The use of hybrid models that combine both has shown promise in improving overall performance.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDistributed Architectures\u003c/strong\u003e: Efficient handling of large-scale IOT data has been made possible through the use of distributed data processing frameworks, including Apache Spark. These frameworks can facilitate real-time analytics, but they may require significant computational resources and infrastructure.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eEdge Computing\u003c/strong\u003e: Edge computing has been the focus of recent studies to reduce latency in real-time analytics. Processing data closer to the source is a way to obtain immediate insights, but this approach can be limited by the processing power available at the edge.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eEnsemble methods\u003c/strong\u003e: A proposal has been made to incorporate ensemble techniques to increase prediction accuracy and robustness. However, these methods add to the complexity and computational demands.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp skip=\"true\"\u003eThe reviewed literature highlights the diverse approaches available for real-time IOMT data processing, each with its own strengths and weaknesses. By integrating the complementary capabilities of CNN and GRU into a distributed architecture, there is an opportunity to enhance these methods. In this paper, to fill this gap, we present a hybrid deep learning model, \u0026quot;CNN-GRU\u0026quot;, based on GRU and CNN for processing IOMT data with a distributed processing environment leveraging Apache Spark and Apache Ignite.\u003c/p\u003e"},{"header":"4- Proposed Model","content":"\u003cp\u003eCurrently, the healthcare system is essential in people\u0026apos;s lives because of the vital role it plays in monitoring people\u0026apos;s health. Processing big data in real time is challenging because of the need for error tolerance and information stability.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eIn this\u0026nbsp;work, we employ a combined method using Apache Spark and Apache Ignite to address the data processing challenges of healthcare systems. A proposed system for data processing and monitoring, in which the data sent by the connected devices is first processed in a distributed database and then stored for real-time analysis, is presented\u003cspan dir=\"RTL\"\u003e.\u003c/span\u003e By combining Apache Ignite and Apache Spark with accurate deep learning algorithms on large-scale medical big data, we aim to achieve a better and faster understanding of diseases, improve disease diagnosis, and enhance patient treatment and prevention. In this study, we utilized an integrated approach involving Apache Spark and Apache Ignite to achieve rapid data access. Additionally, deep learning models including CNN and GRU were employed, as detailed in the following section.. Figure 9 illustrates the structure of the system we have developed.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-1-\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;Hybrid CNN-GRU Algorithm\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis algorithm is designed for real-time medical Big Data (IOMT) processing by integrating Convolutional Neural Networks (CNN) for spatial feature extraction with Gated Recursive Units (GRU) for time sequence processing. Apache Spark\u0026nbsp;is utilized for distributed data processing, while Apache Ignite is used for real-time data processing and in-memory storage to accelerate data access in this algorithm.\u003c/p\u003e\n\u003cp\u003eThe CNN-GRU hybrid algorithm with Spark Apache and Apache Ignite is shown in Figure 10, which is described in the following step-by-step review.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-2- Dataset Description\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, we utilize a dataset about heart disease, which encompasses a range of patient attributes including age, sex, blood pressure, cholesterol levels, and additional factors relevant to cardiac health. The dataset comprises 14 distinct features. Our objective is to employ this dataset to estimate the probability of heart disease occurrence in individuals. Through the application of deep learning techniques to these data, it is feasible to uncover patterns linked to heart disease risk, ultimately aiding in enhanced diagnostic accuracy and preventive strategies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4-3- Data preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this section, the data undergoes processing to make it suitable for application in deep learning models. Within this study, preprocessing played a crucial and significant role in guaranteeing that large-scale medical data obtained from IoT devices were clean, consistent, and adequately prepared for utilization in deep learning algorithms. Preprocessing techniques are needed to ensure the compatibility of the algorithms used with medical data that comes in different forms, such as image data, classified features, and unstructured text. This article discusses these techniques in detail. The preprocessing steps are as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4-3-1- Drop faulty data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAt this phase, data that are incomplete or invalid and could adversely affect the model\u0026rsquo;s effectiveness are detected and eliminated. Data containing missing or incorrect values should be deleted or corrected to improve the model\u0026apos;s prediction accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e \u003c/span\u003e\u003c/strong\u003e\u003cstrong\u003e4-3-2- Grouping features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAt this stage, the data features are categorized based on their type to facilitate easier processing \u0026nbsp;in subsequent steps. This categorization allows us to apply appropriate strategies for each data type.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4-3-3- Numerical features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNumeric characteristics,\u0026nbsp;including age, blood pressure, cholesterol levels, and heart rate, are numerical. For these features to be processed correctly in the model, scaling is necessary. The min\u0026ndash;max scaler technique is used for scaling, and the data are mapped to a range between 0 and 1.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4-3-4- Categorical features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCategorical variables such as age, blood pressure, cholesterol levels, and heart rate are transformed into numerical representations using label encoding methods. This preprocessing step facilitates the integration of these features into deep learning models, thereby enhancing the models\u0026rsquo; predictive accuracy and overall performance in heart disease detection.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-4-\u003c/strong\u003e \u003cstrong\u003eSystem Setup\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this research, a hybrid architecture was employed to optimize the real-time processing of medical big data in the IOT. The system used for implementation is configured as follows in Table 2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u0026nbsp;\u003c/strong\u003e Hardware and Software Configurations\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img1758793737.png\" alt=\"image\" width=\"627\" height=\"241\"\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003eIn this work, four client systems are utilized within a cluster network for real-time data processing. Clients receive their data in a distributed manner and perform processing operations on it. The client systems used for implementation are configured as follows in Table 3.\u0026nbsp;\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e\u0026nbsp; Four client hardware and software configurations\u003cstrong\u003e\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img175879373788.png\" alt=\"image\" width=\"627\" height=\"241\"\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-5- Configuration Models\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp skip=\"true\"\u003eIn this study, several deep learning models were configured to diagnose diseases from big medical data collected through IOT devices. The models were carefully selected and tuned to ensure optimal performance in real-time medical data processing. The configuration of these models included setting appropriate meta-parameters, feature selection, and training methods to maximize accuracy and efficiency in disease diagnosis. The details of the models are presented in Table 4.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u0026nbsp;\u003cstrong\u003eTable 4.\u0026nbsp;\u003c/strong\u003e Model setting specifications\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img175879373742.png\" alt=\"image\" width=\"667\" height=\"846\"\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-6- Data Splitting\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, three distinct databases were employed for disease detection, with the dataset divided into 70% for training purposes and 30% for testing. The data from these sources were separated into training and testing subsets to enable an effective assessment of the model\u0026rsquo;s performance under practical conditions. By allocating 70% of the data for model training and reserving 30% for evaluation, the approach provides sufficient information for the model to learn while maintaining a dedicated portion to test its ability to generalize to new, unseen cases.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-7- Performance metric\u003c/strong\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003eSeveral evaluation criteria can be used to measure and scrutinize the performance of a given classification algorithm. Thus, choosing the appropriate criteria depends on the problem we are dealing with.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eIn some cases, accuracy may be the proper choice, but in other cases, recall or precision may be more appropriate for this purpose. Since we are dealing with medical cases, we can use recall (Sensitivity or True Positive Rate) as a performance metric to choose our classifier. The evaluation metrics of accuracy, recall, precision, and F1 score are used. The performance of the disease detection system is assessed via formulas 1 to 4.\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img1758793785.png\" width=\"1128\" height=\"594\"\u003e\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u0026nbsp;\u003cstrong\u003e4-8- Building the Hybrid Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis section presents the architecture of a hybrid model grounded in deep learning principles. Fundamental deep learning models are integrated with Apache Spark and Apache Ignite to enhance the efficiency of real-time processing for large-scale medical data within the IOMT framework. This architecture makes use of both spatial and temporal analysis capabilities to ensure efficient large-scale data processing. The proposed hybrid model, which combines the strengths of CNNs and GRUs, can create a powerful architecture that can handle various types of big medical data. \u0026nbsp;CNN layers are utilized for processing and extracting features from data, while GRU layers focus on interpreting sequential information from sensor readings and patient health records. Apache Spark stands out as a leading framework for data processing. Its distributed computing architecture enables efficient and scalable management of extensive medical datasets. Spark facilitates operations including data preprocessing, distributed storage, and the parallel execution of machine learning tasks, such as training models based on CNN and GRU architectures. Spark\u0026rsquo;s MLlib is utilized for preliminary model training, data segmentation, and the real-time analysis of large-scale medical data streams. Apache Ignite\u0026apos;s in-memory computing capabilities help speed up data access and reduce latency in data retrieval and processing. Ignite and Spark are seamlessly integrated to provide low-latency storage and cache medical big data in memory for faster computation. The architecture of the hybrid model is outlined as follows:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eInput Layer:\u003c/strong\u003e The input layer receives diverse medical data streams: spatial data from medical imaging systems and temporal data from IoT-enabled wearables and sensors.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCNN Module:\u003c/strong\u003e The spatial data are processed through multiple CNN layers to extract high-dimensional features, identifying critical medical conditions such as tumors or fractures in images.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGRU Module:\u003c/strong\u003e Simultaneously, the sequential time series data from the sensors are passed through the GRU layers for temporal pattern recognition.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDistributed Processing with Spark and Ignite:\u003c/strong\u003e Spark orchestrates the data flow between the CNN and GRU models, parallelizing operations. Ignite caches intermediate results, accelerating real-time processing and reducing I/O bottlenecks.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eOutput Layer:\u003c/strong\u003e The final layer integrates the results from the CNN and GRU to generate predictions, such as disease detection or patient health monitoring outcomes.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe purpose of this architecture is to gather the best features of each model and optimize the spatial and temporal aspects of large medical datasets for real-time processing. Compared with individual models, this combined approach offers superior performance, increased diagnostic accuracy, and timely insight to healthcare professionals. The proposed model is evaluated in Fig. 11.\u0026nbsp;\u003c/p\u003e\n\u003cp skip=\"true\"\u003e\u003cstrong\u003e4-9-Advantages of the Proposed Hybrid Model\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eReal-time Performance:\u003c/strong\u003e By utilizing Spark for large-scale distributed data processing and ignoring in-memory data storage, the system is capable of processing complex medical datasets with near-real-time performance.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eScalability:\u003c/strong\u003e Horizontal scaling of the model is accomplished by incorporating additional nodes into Spark clusters, allowing for efficient management of growing amounts of medical big data.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eAccuracy:\u003c/strong\u003e The combination of a CNN and a GRU for time series data enhances diagnostic accuracy by considering both spatial and temporal patterns.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLow Latency:\u003c/strong\u003e Low-latency access to frequently accessed data is guaranteed by ignoring in-memory capabilities, which enables faster processing and decision-making in critical healthcare applications.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp skip=\"true\"\u003eBy integrating the deep learning strengths of CNN and GRU with the distributed computing capabilities of Apache Spark and Apache Ignite, the proposed hybrid model efficiently manages the real-time processing of large-scale medical data. The efficient and real-time processing of medical big data is crucial for improving patient outcomes in modern healthcare systems, which is why this architecture is well-suited for these patients.\u0026nbsp;\u003c/p\u003e"},{"header":"5-Result and discussion","content":"\u003ch3\u003e\u003cstrong\u003e5-1-Result\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eIn this section, the performances of the CNN, GRU, LSTM, and BILSTM deep learning models, as well as the proposed model, are reported in Tables 5 to 9. Additionally, general diagrams of the models\u0026apos; efficiency evaluation and execution time are shown.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5.\u0026nbsp;\u003c/strong\u003e Results of the CNN\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\" style=\"width: 10px;\"\u003e\u003cstrong\u003eCNN\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u003cstrong\u003eClass\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of spark (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Ignite (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Hybrid (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;0\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e99.79\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.2\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e90.79\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.49\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e11.01\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e7.51\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e9.68\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.38\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.97\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e90.38\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.68\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003eaverage\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.59\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.58\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e90.59\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e90.58\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe performance results of the CNN model in disease diagnosis classification of two different classes are shown in Table 5. The accuracy, precision, recall, and F1 score columns are used to evaluate the performance quality of the model. The average accuracy and recall criteria for the whole model are 90.59%, and the precision and F1 score criteria are 90.58%, which are close to 90.59%.\u003cspan dir=\"RTL\"\u003e\u0026nbsp;\u003c/span\u003eAdditionally, the execution times of the model in three different environments, Spark, Ignite, and a combination of Ignite-Spark, are 11.01 seconds, 7.51 seconds, and 9.68 seconds, respectively. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 6.\u0026nbsp;\u003c/strong\u003e Results of the GRU\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\" style=\"width: 10px;\"\u003e\u003cstrong\u003eGRU\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u003cstrong\u003eClass\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of spark (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Ignite (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Hybrid (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;0\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e100\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e97.99\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e24.38\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e23.13\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e24.06\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e100\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e96.3\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e100\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.11\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003eaverage\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.15\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.05\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 6 shows the performance results of the GRU model in the classification of two different classes of disease diagnosis. Accuracy, precision, recall, and F1 score columns are used to evaluate the performance quality of the model. The average accuracy and recall criterion for the whole model is 98.03%, the precision criterion is 98.15%, and the F1 score is 98.05%. The model\u0026apos;s execution time was evaluated in three different environments: Spark, Ignite, and the combined Ignite-Spark, with times of 24.38, 23.13, and 24.06 seconds, respectively. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 7.\u0026nbsp;\u003c/strong\u003e Results of LSTM\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\" style=\"width: 10px;\"\u003e\u003cstrong\u003eLSTM\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u003cstrong\u003eClass\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of spark (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Ignite (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12px;\"\u003e\u003cstrong\u003eExecution Time of Hybrid (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;0\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e99.76\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e79.72\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\" style=\"width: 12px;\"\u003e92.76\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\u0026nbsp;1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.08\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.08\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e98.08\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.08\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003eaverage\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e98.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e98.05\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 7 shows the performance results of the LSTM model in the classification of two different classes of disease diagnosis. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average of all the requirements for the LSTM model is equal to 98.05%. The running time of the model has been checked in three different environments: Spark, Ignite, and combined Ignite-Spark, which are 99.76, 79.72, and 92.76 seconds, respectively.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 8.\u003c/strong\u003e\u0026nbsp; Results of the BILSTM\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\"\u003e\u003cstrong\u003eBILSTM\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eClass\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of spark (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of Ignite (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of Hybrid (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u0026nbsp;0\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e141.91\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e148.32\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e153.5\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u0026nbsp;1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.15\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.15\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.15\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.15\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003eaverage\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e96.1\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 8 shows the performance results of the BILSTM model in the classification of two different classes of disease diagnosis. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average of all the requirements for the BILSTM model is 96.10%. The running time of the model has been checked in three different environments: Spark, Ignite, and combined Ignite-Spark, which are 141.91, 148.32 seconds, and 153.50 seconds, respectively. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 9.\u003c/strong\u003e\u0026nbsp; Results of the proposed model (CNN+GRU)\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\"\u003e\u003cstrong\u003eBILSTM\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eClass\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of spark (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of Ignite (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cstrong\u003eExecution Time of Hybrid (s)\u003c/strong\u003e\u003cbr\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u0026nbsp;0\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e100.00\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e98.03\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.00\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e18.84\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e18.93\u003cbr\u003e\u003c/td\u003e\n \u003ctd rowspan=\"3\" valign=\"top\"\u003e18.18\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u0026nbsp;1\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e100.00\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e98.11\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e100.00\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.05\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003eaverage\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.01\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.06\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.01\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e99.03\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 9 displays the performance outcomes of the proposed model in the classification of two distinct disease diagnosis classes. The accuracy, precision, recall, and F1 score criteria are used to evaluate the performance quality of the model. The average accuracy and recall criterion for the proposed model are 99.01% and 99.06%, respectively, with a F1 score of 99.03%. The execution time of the model has been checked in three different environments: Spark, Ignite, and the combination of Ignite-Spark, which are 18.84, 18.93 seconds, and 18.18 seconds, respectively. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 12 shows an evaluation of the deep learning models with four main criteria. As shown in Figure 11, the CNN and BILSTM methods are less efficient than the other methods. The GRU and LSTM models exhibit similar performances, whereas the CNN-GRU (Proposed Model) outperforms the different models.\u003c/p\u003e\n\u003cp\u003eFigure 12 illustrates the performance of various deep learning models, demonstrating the clear advantage of the proposed approach. The times of the calculations and execution of the models are very different. Figure 13 shows that the CNN and GRU models take less time than the other methods do in different Spark, Ignite, and hybrid environments. Consequently, a hybrid approach integrating both CNN and GRU architectures was proposed, demonstrating greater efficiency relative to alternative techniques. When explicitly compared to the GRU model alone, this combined model exhibits superior efficiency percentages. Compared with the GRU model, the execution time of the proposed hybrid model has been improved by nearly 13%, 12%, and 19%, respectively, in different Spark, Ignite, and Ignite-Spark environments, which shows the high efficiency of computing time and faster execution. Next, the proposed model (CNN-GRU) is compared with other methods in Table 10. It can be seen that the proposed model achieves a higher accuracy compared to the different models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 10.\u003c/strong\u003e\u0026nbsp; Evaluation of the proposed model with other methods.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"87%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003eRef\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\u003cstrong\u003eYear\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\u003cstrong\u003eMethod\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[105]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2020\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eDeep Neural Network\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e98.10\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[106]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2020\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eCNN + LASSO\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e85.70\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[107]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2021\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eCNN\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e97.00\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[108]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2021\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eDL-based Classifier\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e94.20\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[109]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eBi-LSTM\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e98.80\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[110]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eEnsemble classifier\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e89.00\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[111]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2022\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eXGBoost with RFE\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e95.60\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[112]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2023\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eRF\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e97.00\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[113]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2023\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eHybrid CNN-LSTM\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e74.20\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[114]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2024\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eOptimized SVM\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e96.00\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003e[115]\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e2024\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eOptimized XGBoost\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e98.40\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 22px;\"\u003e\u003cstrong\u003eCurrent research\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003eProposed model (CNN-GRU)\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e99.01\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e5-2-Discussion\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this section, we provide a comparative analysis of the execution times observed across five distinct models\u0026nbsp;using three processing systems (Spark, Ignite, and Hybrid). These models include CNN, GRU, LSTM, BILSTM, and the Proposed Model, which are typically used for data analysis, particularly in deep learning tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCNN:\u003c/strong\u003e In the CNN model, the execution time with Ignite (7.51 seconds) is clearly faster than Spark (11.01 seconds). This time difference is primarily due to Ignite\u0026rsquo;s in-memory data storage, which enables faster processing for smaller or real-time data processing tasks. However, the execution time using the Hybrid method (9.68 seconds) is quite close to Ignite, indicating that the hybrid system performs almost as efficiently as Ignite, with a slight overhead. This suggests that combining both systems results in optimized performance despite some additional coordination overhead.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGRU:\u0026nbsp;\u003c/strong\u003eFor the GRU model, the execution times for Spark (24.38 seconds) and Ignite (23.13 seconds) are pretty close. Here, Ignite remains slightly faster than Spark, likely due to its quicker data retrieval from in-memory storage. The execution time for the Hybrid method (24.06 seconds) is somewhat longer than Ignite, but still comparable. This increase in time is attributed to the coordination overhead when combining the two systems. However, when both systems work together in parallel processing, the benefits often outweigh the marginal increase in time.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLSTM:\u003c/strong\u003e In the LSTM model, which typically requires processing longer and more complex data sequences, ignite (79.72 seconds) is faster than Spark (99.76 seconds). This difference suggests that Ignite is more efficient at handling memory-intensive tasks and real-time data processing. However, the execution time with the Hybrid method (92.76 seconds) is slightly slower than Ignite. This is due to the additional overhead from coordinating and synchronizing tasks between the two systems. Nonetheless, this slight increase in time is still justified by the combined advantages of both systems in handling large-scale and complex data processing tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBILSTM:\u003c/strong\u003e In the BILSTM model, which involves processing complex and lengthy data sequences, Spark (141.91 seconds) is faster than Ignite (148.32 seconds). The time difference between these two systems reflects the increased computational requirements of the model. However, the execution time with the Hybrid method (153.5 seconds) is the highest among the three, reflecting the additional coordination and resource synchronization overhead when combining both systems. Although this results in a slightly slower execution time, the hybrid method\u0026apos;s ability to handle both extensive data processing and real-time tasks more flexibly still makes it a valuable option for complex models like BILSTM.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProposed Model:\u003c/strong\u003e In the Proposed Model, the execution times for both Spark (18.84 seconds) and Ignite (18.93 seconds) are almost identical, showing that both systems perform similarly in this case. However, the execution time for the Hybrid method (18.18 seconds) is slightly faster than both, which indicates that the combination of both systems in this model has been particularly optimized to minimize overhead and improve execution time. The Hybrid approach leverages the advantages of Spark for handling extensive data processing tasks alongside Ignite\u0026rsquo;s capabilities for real-time, in-memory computations. By integrating these technologies, the hybrid system demonstrates notable efficiency in scenarios demanding both large-scale data handling and rapid computational performance, such as with LSTM and BILSTM models. This method contributes to decreased execution times in specific instances while also enhancing the system\u0026rsquo;s flexibility and scalability when managing substantial and intricate datasets. The Hybrid method is highly scalable, combining the ability of Spark to handle massive datasets with Ignite\u0026rsquo;s real-time processing capabilities. As a result, the hybrid system offers significant benefits in situations that demand both extensive data processing capabilities and real-time analytical performance. The flexibility to switch between these two systems according to the task at hand allows for optimal performance across different use cases, ensuring that both data volume and processing speed are addressed efficiently. By using a hybrid approach, the system can better manage resources by utilizing Spark for batch processing and Ignite for in-memory real-time processing. This method leads to more optimized resource utilization, as Spark can handle large-scale computations, while Ignite provides fast access to data stored in memory for real-time tasks. This method guarantees efficient resource allocation, resulting in accelerated execution and enhanced system performance, particularly for complex models. Analysis of execution times and evaluation of the advantages provided by the hybrid strategy indicate that, for the majority of models, the Hybrid approach surpasses the performance of using either Spark or Ignite individually, owing to its capacity to leverage the strengths of both platforms. While in some models like BILSTM, the hybrid method results in slightly higher execution times, the overall performance improvement and increased flexibility in handling complex tasks justify this increase in time.\u003c/p\u003e"},{"header":"6- Conclusion","content":"\u003cp\u003eIn this study, we employ Apache Spark and Apache Ignite in combination with advanced deep learning techniques, recognized for their speed and high efficiency in real-time analysis of large-scale medical datasets. We have presented a hybrid model that integrates CNN and GRU architectures, developed explicitly for managing real-time data generated by IOMT devices in healthcare and patient management systems. Our model efficiently manages the intricate nature of IOMT data by employing a CNN to extract relevant features and a GRU to capture temporal relationships, both of which are essential for processing high-dimensional and sequential data characteristics. The scalability and speed of model processing can be significantly improved by implementing a hybrid distributed architecture via Apache Spark and Ignite. This architecture is designed to enable efficient data capture, real-time analysis, and seamless integration into existing healthcare systems, leading to faster and timelier decision-making by physicians and improved patient outcomes. Our experimental results show that the CNN-GRU hybrid model outperforms previous conventional approaches in terms of speed, accuracy, and processing latency, highlighting its suitability for practical applications. Furthermore, the model exhibits strong adaptability across various IOMT environments, as evidenced by its robustness when working with data of different qualities and formats.\u003c/p\u003e\u003cp\u003eThe objective of this approach was to minimize the processing time involved in real-time handling of large-scale medical data during data retrieval tasks, while simultaneously enhancing execution speed and overall efficiency within systems designed for disease diagnosis and prediction. By combining the deep learning-based CNN and GRU models, we enhance the real-time processing capabilities and accuracy of medical data analysis in healthcare systems. Compared to previous models, our hybrid approach demonstrated improved performance in terms of processing speed, operational efficiency, and accuracy. These improvements suggest that this method has significant potential to advance outcomes in healthcare systems.\u003c/p\u003e\u003cp\u003eBased on these findings, we combined a CNN and GRU into a hybrid \u003cb\u003eCNN-GRU\u003c/b\u003e model, which achieved improved results in both time and accuracy. The experimental findings highlight the proposed system\u0026rsquo;s capability to efficiently handle and forecast substantial volumes of real-time medical data originating from various distributed diseases. Assessment of the developed CNN-GRU model demonstrated outstanding results, achieving an accuracy of 99.01%, a recall of 99.01%, a precision of 99.06%, and an F1 score of 99.03%. The model\u0026rsquo;s performance was further validated by measuring execution times across three different platforms: Spark, Ignite, and their combination, yielding times of 18.84 seconds, 18.93 seconds, and 19.28 seconds, respectively, thereby confirming its effectiveness and efficiency in diverse computational environments. The objective of this approach was to minimize the time needed for real-time processing of large-scale medical data during data retrieval tasks, while simultaneously enhancing execution speed and overall efficiency in systems for disease diagnosis and prediction. By integrating deep learning architectures such as CNN and GRU, we have strengthened both the real-time processing capacity and the accuracy of medical data analysis within healthcare frameworks. When compared to earlier models, our hybrid technique exhibited notable improvements in processing speed, efficiency, and predictive accuracy, positioning it as a promising strategy for advancing healthcare outcomes. This study represents a substantial contribution to the field of real-time IOMT data processing.\u003c/p\u003e\u003cp\u003eDespite the surprising progress shown in this study, future work could be done by incorporating other methods, or this hybrid method can be expanded and developed into a more robust, scalable, and secure solution for real-time medical data processing by combining machine learning algorithms alongside CNN and GRU deep learning models. We can improve the system via advanced machine learning techniques. These include ensemble learning, decision trees, and SVMs. The CNN-GRU model can increase accuracy and adaptability. These advancements will increase disease diagnosis efficiency and accuracy. They also improve healthcare management, patient health, and disease control.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthors:A. Dr. Erfaneh Noroozi)Corresponding Author(B. Phd student- Mohamad Karamkish Zahooki C. Dr. Mehdi Hosseinzadeh\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDeclaration of Competing Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors confirm\u0026nbsp;there are no conflicts of interest associated with this\u0026nbsp;research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll datasets examined and employed in this study are available from the authors upon reasonable request. The primary dataset analyzed is openly accessible via Kaggle at the following URL: https://www.kaggle.com/datasets/johnsmith88/heart-disease-dataset.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKashani, M.H., et al., \u003cem\u003eA systematic review of IoT in healthcare: Applications, techniques, and trends.\u003c/em\u003e 2021. \u003cstrong\u003e192\u003c/strong\u003e: p. 103164.\u003c/li\u003e\n\u003cli\u003eShrotriya, L., et al., \u003cem\u003eApache Spark in healthcare: Advancing data-driven innovations and better patient care.\u003c/em\u003e 2023. \u003cstrong\u003e14\u003c/strong\u003e(6).\u003c/li\u003e\n\u003cli\u003eHashemi, S., et al., \u003cem\u003eService and energy management in fog computing: A taxonomy approach and future directions.\u003c/em\u003e 2024. \u003cstrong\u003e12\u003c/strong\u003e(1): p. 15-38.\u003c/li\u003e\n\u003cli\u003eHashemi, S.M., et al., \u003cem\u003eGwo-sa: Gray wolf optimization algorithm for service activation management in fog computing.\u003c/em\u003e 2022. \u003cstrong\u003e10\u003c/strong\u003e: p. 107846-107863.\u003c/li\u003e\n\u003cli\u003ePlageras, A.P., et al. \u003cem\u003eEfficient large-scale medical data (ehealth big data) analytics in the Internet of Things\u003c/em\u003e, in \u003cem\u003e2017, IEEE 19th Conference on Business Informatics (CBI)\u003c/em\u003e. 2017. IEEE.\u003c/li\u003e\n\u003cli\u003eYadav, B., et al., \u003cem\u003eInternet of Medical Things (IOMT) for Chronic Disease Management.\u003c/em\u003e 2024. \u003cstrong\u003e13\u003c/strong\u003e(6): p. 2278-0181.\u003c/li\u003e\n\u003cli\u003eHashemi, S.M., et al., \u003cem\u003eA new approach for service activation management in fog computing using Cat Swarm Optimization algorithm.\u003c/em\u003e 2024: p. 1-36.\u003c/li\u003e\n\u003cli\u003eME, B.K., \u003cem\u003eAutonomous Intervention Medical Tools\u003c/em\u003e. 2024, Worcester Polytechnic Institute.\u003c/li\u003e\n\u003cli\u003eAhmed, S.F., et al., \u003cem\u003eInsights into Internet of Medical Things (IoMT): Data fusion, security issues and potential solutions.\u003c/em\u003e 2024. \u003cstrong\u003e102\u003c/strong\u003e: p. 102060.\u003c/li\u003e\n\u003cli\u003eEl-Saleh, A.A., et al., \u003cem\u003eThe Internet of Medical Things (IoMT): opportunities and challenges.\u003c/em\u003e 2024: p. 1-18.\u003c/li\u003e\n\u003cli\u003ePradhan, B., S. Bhattacharyya, and K.J.J.o.h.e. Pal, \u003cem\u003eIoT‐based applications in healthcare devices.\u003c/em\u003e 2021. \u003cstrong\u003e2021\u003c/strong\u003e(1): p. 6632599.\u003c/li\u003e\n\u003cli\u003eAmiri, Z., et al., \u003cem\u003eThe deep learning applications in IoT-based bio-and medical informatics: a systematic literature review.\u003c/em\u003e 2024. \u003cstrong\u003e36\u003c/strong\u003e(11): p. 5757-5797.\u003c/li\u003e\n\u003cli\u003eMilasus, E., \u003cem\u003eDesign and Implementation of Distributed System Application for Gathering, Harmonizing, and Distributing Data Acquired from Different Data Sources\u003c/em\u003e. 2024, Technische Universit\u0026auml;t Wien.\u003c/li\u003e\n\u003cli\u003eArshad, H., et al., \u003cem\u003eTrends and Challenges in harnessing big data intelligence for health care transformation.\u003c/em\u003e 2024: p. 220-240.\u003c/li\u003e\n\u003cli\u003eManogaran, G., et al., \u003cem\u003eA new architecture of Internet of Things and big data ecosystem for secured intelligent healthcare monitoring and alerting system.\u003c/em\u003e 2018. \u003cstrong\u003e82\u003c/strong\u003e: p. 375-387.\u003c/li\u003e\n\u003cli\u003eAminizadeh, S., et al., \u003cem\u003eOpportunities and challenges of artificial intelligence and distributed systems to improve the quality of healthcare service.\u003c/em\u003e 2024. \u003cstrong\u003e149\u003c/strong\u003e: p. 102779.\u003c/li\u003e\n\u003cli\u003eTitus, A., et al., \u003cem\u003eE-Healthcare Data Management Using Machine Learning and IoT\u003c/em\u003e, in \u003cem\u003eIoT and ML for Information Management: A Smart Healthcare Perspective\u003c/em\u003e. 2024, Springer. p. 167-199.\u003c/li\u003e\n\u003cli\u003eNazari, E., M.H. Shahriari, and H.J.F. H.I. Tabesh, \u003cem\u003eBig Data analysis in healthcare: Apache Hadoop, Apache Spark, and Apache Flink.\u003c/em\u003e 2019. \u003cstrong\u003e8\u003c/strong\u003e(1): p. 14.\u003c/li\u003e\n\u003cli\u003eBasco, J.A. and N. Senthilkumar. \u003cem\u003eReal-time analysis of healthcare using big data analytics\u003c/em\u003e. In \u003cem\u003eIOP Conference Series: Materials Science and Engineering\u003c/em\u003e. 2017. IOP Publishing.\u003c/li\u003e\n\u003cli\u003eRay, P.P.J.F.C. and I. Journal, \u003cem\u003eA survey of IoT cloud platforms.\u003c/em\u003e 2016. \u003cstrong\u003e1\u003c/strong\u003e(1-2): p. 35-46.\u003c/li\u003e\n\u003cli\u003eWan, K. and V. Alagar. \u003cem\u003eCharacteristics and Classification of Big Data in the Healthcare Sector\u003c/em\u003e. In \u003cem\u003e2016, the 12th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD)\u003c/em\u003e. 2016. IEEE.\u003c/li\u003e\n\u003cli\u003eBaloch, L., et al., \u003cem\u003eA review of big data trends and challenges in healthcare.\u003c/em\u003e 2023. \u003cstrong\u003e14\u003c/strong\u003e(6): p. 1320-1333.\u003c/li\u003e\n\u003cli\u003eHassan, S., et al., \u003cem\u003eBig data and predictive analytics in healthcare in Bangladesh: regulatory challenges.\u003c/em\u003e 2021. \u003cstrong\u003e7\u003c/strong\u003e(6).\u003c/li\u003e\n\u003cli\u003eAhmed, V., et al., \u003cem\u003eThe future of big data in facilities management: opportunities and challenges.\u003c/em\u003e 2017. \u003cstrong\u003e35\u003c/strong\u003e(13/14): p. 725-745.\u003c/li\u003e\n\u003cli\u003eKhatiwada, P., et al., \u003cem\u003ePatient-Generated Health Data (PGHD): Understanding, Requirements, Challenges, and Existing Techniques for Data Security and Privacy.\u003c/em\u003e 2024. \u003cstrong\u003e14\u003c/strong\u003e(3): p. 282.\u003c/li\u003e\n\u003cli\u003eYang, X., et al., \u003cem\u003eBiomedical big data technologies, applications, and challenges for precision medicine: A review.\u003c/em\u003e 2024. \u003cstrong\u003e8\u003c/strong\u003e(1): p. 2300163.\u003c/li\u003e\n\u003cli\u003eWu, P.-Y., et al., \u003cem\u003e\u0026ndash;Omic and electronic health record big data analytics for precision medicine.\u003c/em\u003e 2016. \u003cstrong\u003e64\u003c/strong\u003e(2): p. 263-273.\u003c/li\u003e\n\u003cli\u003ePradyumna, G., et al., \u003cem\u003eEmpowering healthcare with IoMT: Evolution, machine learning integration, security, and interoperability challenges.\u003c/em\u003e 2024.\u003c/li\u003e\n\u003cli\u003eSingh, R.K.J.D.L.P., \u003cem\u003eDeveloping a big data analytics platform using Apache Hadoop Ecosystem for delivering big data services in libraries.\u003c/em\u003e 2024. \u003cstrong\u003e40\u003c/strong\u003e(2): p. 160-186.\u003c/li\u003e\n\u003cli\u003eCena, F., et al., \u003cem\u003eIntroduction to the special issue on big personal data in interactive intelligent systems.\u003c/em\u003e 2017. \u003cstrong\u003e7\u003c/strong\u003e(2): p. 1-5.\u003c/li\u003e\n\u003cli\u003ePramanik, P.K.D., et al., \u003cem\u003eHealthcare big data: A comprehensive overview.\u003c/em\u003e 2022: p. 119-147.\u003c/li\u003e\n\u003cli\u003eViceconti, M., et al., \u003cem\u003eBig data, big knowledge: big data for personalized healthcare.\u003c/em\u003e 2015. \u003cstrong\u003e19\u003c/strong\u003e(4): p. 1209-1215.\u003c/li\u003e\n\u003cli\u003eEl-Gayar, O. and P. Timsina. \u003cem\u003eOpportunities for business intelligence and big data analytics in evidence-based medicine\u003c/em\u003e. In \u003cem\u003e2014, the 47th Hawaii International Conference on System Sciences\u003c/em\u003e. 2014. IEEE.\u003c/li\u003e\n\u003cli\u003eLuo, J., et al., \u003cem\u003eBig data application in biomedical research and health care: a literature review.\u003c/em\u003e 2016. \u003cstrong\u003e8\u003c/strong\u003e: p. BII. S31559.\u003c/li\u003e\n\u003cli\u003eHosen, M.S., et al., \u003cem\u003eData-Driven Decision Making: Advanced Database Systems for Business Intelligence.\u003c/em\u003e 2024: p. 687\u0026ndash;704-687\u0026ndash;704.\u003c/li\u003e\n\u003cli\u003eSun, Y., et al., \u003cem\u003eA Real-time Associative Feature-based CRM and ERP Integration Model for SMEs.\u003c/em\u003e 2024: p. 1-19.\u003c/li\u003e\n\u003cli\u003eAhamad, S., et al., \u003cem\u003eApollo Hospital\u0026apos;s Proposed Use of Big Data Healthcare Analytics\u003c/em\u003e, in \u003cem\u003eBig Data Analytics Techniques for Market Intelligence\u003c/em\u003e. 2024, IGI Global. p. 312-328.\u003c/li\u003e\n\u003cli\u003eJevinger, \u0026Aring;., et al., \u003cem\u003eArtificial intelligence for improving public transport: a mapping study.\u003c/em\u003e 2024. \u003cstrong\u003e16\u003c/strong\u003e(1): p. 99-158.\u003c/li\u003e\n\u003cli\u003eSalemi, A. and H. Zamani. \u003cem\u003eTowards a search engine for machines: Unified ranking for multiple retrieval-augmented large language models\u003c/em\u003e. In \u003cem\u003eProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval\u003c/em\u003e. 2024.\u003c/li\u003e\n\u003cli\u003eChowdhury, A.T., et al., \u003cem\u003eApplication of Big Data in Infectious Disease Surveillance: Contemporary Challenges and Solutions\u003c/em\u003e, in \u003cem\u003eSurveillance, Prevention, and Control of Infectious Diseases: An AI Perspective\u003c/em\u003e. 2024, Springer. p. 51-71.\u003c/li\u003e\n\u003cli\u003eHashemi, S.M., et al., \u003cem\u003eEnergy‐aware resource management in fog computing for IoT applications: A review, taxonomy, and future directions.\u003c/em\u003e 2024. \u003cstrong\u003e54\u003c/strong\u003e(2): p. 109-148.\u003c/li\u003e\n\u003cli\u003eTheodorakopoulos, L., A. Theodoropoulou, and Y.J.E. Stamatiou, \u003cem\u003eA State-of-the-Art Review in Big Data Management Engineering: Real-Life Case Studies, Challenges, and Future Research Directions.\u003c/em\u003e 2024. \u003cstrong\u003e5\u003c/strong\u003e(3): p. 1266-1297.\u003c/li\u003e\n\u003cli\u003eYao, Y., et al., \u003cem\u003eNew scheduling algorithms for improving performance and resource utilization in Hadoop YARN clusters.\u003c/em\u003e 2019. \u003cstrong\u003e9\u003c/strong\u003e(3): p. 1158-1171.\u003c/li\u003e\n\u003cli\u003eSalloum, S., et al., \u003cem\u003eBig data analytics on Apache Spark.\u003c/em\u003e 2016. \u003cstrong\u003e1\u003c/strong\u003e: p. 145-164.\u003c/li\u003e\n\u003cli\u003ePonnusamy, S. and P.J.I.A. Gupta, \u003cem\u003eScalable Data Partitioning Techniques for Distributed Data Processing in Cloud Environments: A Review.\u003c/em\u003e 2024.\u003c/li\u003e\n\u003cli\u003eChennuri, S., K.R.J.I.J.O.C.E. Kotha, and TECHNOLOGY, \u003cem\u003eDATA ENGINEERING FOR REAL-TIME ANALYTICS: TURNING DATA INTO INSTANT INSIGHTS.\u003c/em\u003e 2024. \u003cstrong\u003e15\u003c/strong\u003e(5): p. 752-763.\u003c/li\u003e\n\u003cli\u003eBalakrishnan, A.J.I.J.o. C.T. and Technology, \u003cem\u003eData Virtualization Architecture Framework Using Multi-Engine Data Platforms for Big Data Analytics and Machine Learning.\u003c/em\u003e 2024. \u003cstrong\u003e72\u003c/strong\u003e(2): p. 82-91.\u003c/li\u003e\n\u003cli\u003eAL-SHARGABI, A.A., \u003cem\u003eBlockchain-IoT Healthcare Applications and Trends: A Review.\u003c/em\u003e 2024.\u003c/li\u003e\n\u003cli\u003eTalia, D. and P. Trunfio. \u003cem\u003eProgramming Tools for High-Performance Data Analysis\u003c/em\u003e. In \u003cem\u003eProceedings of the 33rd International Symposium on High-Performance Parallel and Distributed Computing\u003c/em\u003e. 2024.\u003c/li\u003e\n\u003cli\u003eAl-Jumaili, A.H.A., et al., \u003cem\u003eParallel power load abnormalities detection using fast density peak clustering with a hybrid canopy-K-means algorithm.\u003c/em\u003e (Preprint): p. 1-26.\u003c/li\u003e\n\u003cli\u003eWu, C.-Y., et al. \u003cem\u003eMachine learning-based real-time task scheduling for Apache Storm\u003c/em\u003e. In \u003cem\u003eSensors and Systems for Space Applications XVII\u003c/em\u003e. 2024. SPIE.\u003c/li\u003e\n\u003cli\u003eAl-Ali, A., et al., \u003cem\u003eRole of IoT technologies in big data management systems: A review and Smart Grid case study.\u003c/em\u003e 2024: p. 101905.\u003c/li\u003e\n\u003cli\u003eHussein, A.e.e.A.J.J.o.I.S., \u003cem\u003eFifty-six big data V\u0026rsquo;s characteristics and proposed strategies to overcome security and privacy challenges (BD2).\u003c/em\u003e 2020. \u003cstrong\u003e11\u003c/strong\u003e(4): p. 304-328.\u003c/li\u003e\n\u003cli\u003eLipovac, I., M.B.J.I.J.o. D.M. Babac, Modelling, and Management, \u003cem\u003eDeveloping a Data Pipeline Solution for Big Data Processing.\u003c/em\u003e 2024. \u003cstrong\u003e16\u003c/strong\u003e(1): p. 1-22.\u003c/li\u003e\n\u003cli\u003eRidzuan, F. and W.M.N.W.J.P.C.S. Zainon, \u003cem\u003eA Review on Data Quality Dimensions for Big Data.\u003c/em\u003e 2024. \u003cstrong\u003e234\u003c/strong\u003e: p. 341-348.\u003c/li\u003e\n\u003cli\u003eAzzam, M.E.A.Y., et al., \u003cem\u003eHow big data features drive financial accounting and firm sustainability in the energy industry.\u003c/em\u003e 2024. \u003cstrong\u003e22\u003c/strong\u003e(1): p. 29-51.\u003c/li\u003e\n\u003cli\u003eBawaneh, R. and A. Jallad. \u003cem\u003eBig Data in Healthcare: A Review on Applications, Technologies, Benefits and Challenges\u003c/em\u003e. in \u003cem\u003e2024 21st International Multi-Conference on Systems, Signals \u0026amp; Devices (SSD)\u003c/em\u003e. 2024. IEEE.\u003c/li\u003e\n\u003cli\u003eOgundipe, D.O.J.I.M.S.R.J., \u003cem\u003eThe impact of big data on healthcare product development: A theoretical and analytical review.\u003c/em\u003e 2024. \u003cstrong\u003e4\u003c/strong\u003e(3): p. 341-360.\u003c/li\u003e\n\u003cli\u003eGuo, C. and J. Chen, \u003cem\u003eBig data analytics in healthcare\u003c/em\u003e, in \u003cem\u003eKnowledge technology and systems: Toward establishing knowledge systems science\u003c/em\u003e. 2023, Springer. p. 27-70.\u003c/li\u003e\n\u003cli\u003eParamesha, M., N.L. Rane, J.J.P.U.M.R.J. Rane, Big Data Analytics, Artificial Intelligence, Machine Learning, Internet of Things, and Blockchain for Enhanced Business Intelligence\u003cem\u003e.\u003c/em\u003e 2024. \u003cstrong\u003e1\u003c/strong\u003e(2): p. 110-133.\u003c/li\u003e\n\u003cli\u003eDang, T., et al., \u003cem\u003eTwo-layer Ensemble of Deep Learning Models for Medical Image Segmentation.\u003c/em\u003e 2024: p. 1-20.\u003c/li\u003e\n\u003cli\u003eKaveh, A.J.S.i.c.i., \u003cem\u003eApplications of Artificial neural networks and machine learning in Civil Engineering.\u003c/em\u003e 2024. \u003cstrong\u003e1168\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eJivisar, H., J. Benson, and A. Kowalski, \u003cem\u003eAutomated detection of compressed and encrypted ransomware data using a three-layer convolutional neural network.\u003c/em\u003e 2024.\u003c/li\u003e\n\u003cli\u003eSeyrek, E.C., M.J.M.T. Uysal, and Applications, \u003cem\u003eA comparative analysis of various activation functions and optimizers in a convolutional neural network for hyperspectral image classification.\u003c/em\u003e 2024. \u003cstrong\u003e83\u003c/strong\u003e(18): p. 53785-53816.\u003c/li\u003e\n\u003cli\u003eZhao, X., et al., \u003cem\u003eA review of convolutional neural networks in computer vision.\u003c/em\u003e 2024. \u003cstrong\u003e57\u003c/strong\u003e(4): p. 99.\u003c/li\u003e\n\u003cli\u003eKishore, J. and S.J.M.I.R. Mukherjee, \u003cem\u003eImpact of autotuned fully connected layers on performance of self-supervised models for image classification.\u003c/em\u003e 2024: p. 1-13.\u003c/li\u003e\n\u003cli\u003eLe, X.-H., et al., \u003cem\u003eDeep neural network-based discharge prediction for upstream hydrological stations: a comparative study.\u003c/em\u003e 2023. \u003cstrong\u003e16\u003c/strong\u003e(4): p. 3113-3124.\u003c/li\u003e\n\u003cli\u003eKavitha, D., T. Adilakshmi, and M.C.J.J.o. E.S. Mohan, \u003cem\u003ePreserving Privacy of IoT Healthcare Data using Differential Privacy and LSTM.\u003c/em\u003e 2024. \u003cstrong\u003e20\u003c/strong\u003e(7s): p. 2483-2492.\u003c/li\u003e\n\u003cli\u003eTzoumpas, K., et al., \u003cem\u003eA data filling methodology for time series based on CNN and (Bi) LSTM neural networks.\u003c/em\u003e 2024. \u003cstrong\u003e12\u003c/strong\u003e: p. 31443-31460.\u003c/li\u003e\n\u003cli\u003eWang, W., et al., \u003cem\u003eA Hybrid ARIMA-LSTM Model for Short-Term Vehicle Speed Prediction.\u003c/em\u003e 2024. \u003cstrong\u003e17\u003c/strong\u003e(15): p. 3736.\u003c/li\u003e\n\u003cli\u003eTucker, A.P., et al., \u003cem\u003eNeural Networks With Gated Recurrent Units Reduce Glucose Forecasting Error Due to Changes in Sensor Location.\u003c/em\u003e 2024. \u003cstrong\u003e18\u003c/strong\u003e(1): p. 124-134.\u003c/li\u003e\n\u003cli\u003eRongali, A.S., et al. \u003cem\u003eChaoG-GRU: Healthcare Transaction Fraud Detection using Chaotic Iterative Gazelle-based Gated Recurrent Unit Technique\u003c/em\u003e. in \u003cem\u003e2024 Third International Conference on Intelligent Techniques in Control, Optimization and Signal Processing (INCOS)\u003c/em\u003e. 2024. IEEE.\u003c/li\u003e\n\u003cli\u003eAlshehri, O.S., O.M. Alshehri, and H. Samma. \u003cem\u003eBlood Glucose Prediction Using RNN, LSTM, and GRU: A Comparative Study\u003c/em\u003e. In \u003cem\u003e2024, the IEEE International Conference on Advanced Systems and Emergent Technologies (IC_ASET)\u003c/em\u003e. 2024. IEEE.\u003c/li\u003e\n\u003cli\u003eTu, B., et al., \u003cem\u003eReal-time prediction of ROP based on GRU-Informer.\u003c/em\u003e 2024. \u003cstrong\u003e14\u003c/strong\u003e(1): p. 2133.\u003c/li\u003e\n\u003cli\u003eZulqarnain, M., et al., \u003cem\u003eAn efficient two-state GRU based on feature attention mechanism for sentiment analysis.\u003c/em\u003e 2024. \u003cstrong\u003e83\u003c/strong\u003e(1): p. 3085-3110.\u003c/li\u003e\n\u003cli\u003eDi Mauro, M., et al. \u003cem\u003eEvaluating Recurrent Neural Networks for the prediction of Multi-Variate time series VoIP metrics\u003c/em\u003e. In \u003cem\u003e2024, the 22nd Mediterranean Communication and Computer Networking Conference (MedComNet)\u003c/em\u003e. 2024. IEEE.\u003c/li\u003e\n\u003cli\u003eAla, A., et al., \u003cem\u003eEnhancing patient information performance in Internet of Things-based smart healthcare system: Hybrid artificial intelligence and optimization approaches.\u003c/em\u003e 2024. \u003cstrong\u003e131\u003c/strong\u003e: p. 107889.\u003c/li\u003e\n\u003cli\u003eGhosh, S., A. Ekbal, and P. Bhattacharyya, \u003cem\u003eNatural language processing and sentiment analysis: perspectives from computational intelligence\u003c/em\u003e, in \u003cem\u003eComputational Intelligence Applications for Text and Sentiment Data Analysis\u003c/em\u003e. 2023, Elsevier. p. 17-47.\u003c/li\u003e\n\u003cli\u003eVishnupriya, G., S. Anusha, and S.J.T.o., E.T.T. Kayikci, \u003cem\u003eAn efficient and secure wearable sensor-based remote healthcare monitoring system using adaptive dilated transformer Bi‐LSTM with gated recurrent unit.\u003c/em\u003e 2024. \u003cstrong\u003e35\u003c/strong\u003e(2): p. e4932.\u003c/li\u003e\n\u003cli\u003eAburass, S., et al., \u003cem\u003eA hybrid machine learning model for classifying gene mutations in cancer using LSTM, BiLSTM, CNN, GRU, and GloVe.\u003c/em\u003e 2024. \u003cstrong\u003e6\u003c/strong\u003e: p. 200110.\u003c/li\u003e\n\u003cli\u003eMienye, I.D., T.G. Swart, and G.J.I. Obaido, \u003cem\u003eRecurrent Neural Networks: A Comprehensive Review of Architectures, Variants, and Applications.\u003c/em\u003e 2024. \u003cstrong\u003e15\u003c/strong\u003e(9): p. 517.\u003c/li\u003e\n\u003cli\u003eWu, Z.J.J.o.C.S. and T. Studies, \u003cem\u003eDeep Learning with Improved Metaheuristic Optimization for Traffic Flow Prediction.\u003c/em\u003e 2024. \u003cstrong\u003e6\u003c/strong\u003e(4): p. 47-53.\u003c/li\u003e\n\u003cli\u003eJerse, G., A.J.A. Marcucci, and Computing, \u003cem\u003eDeep Learning LSTM-based approaches for 10.7 cm solar radio flux forecasting up to 45 days.\u003c/em\u003e 2024. \u003cstrong\u003e46\u003c/strong\u003e: p. 100786.\u003c/li\u003e\n\u003cli\u003eLi, X. and S.J.M. Liu, \u003cem\u003ePredicting 30-day hospital readmission in medicare patients: Insights from an LSTM deep learning model.\u003c/em\u003e 2024: p. 2024.09. 08.24313212.\u003c/li\u003e\n\u003cli\u003eBusst, M.M.A., et al., \u003cem\u003eEnsemble BiLSTM: A Novel Approach for Aspect Extraction From Online Text.\u003c/em\u003e 2024.\u003c/li\u003e\n\u003cli\u003eWankhade, M., et al., \u003cem\u003eCBMAFM: CNN-BiLSTM multi-attention fusion mechanism for sentiment classification.\u003c/em\u003e 2024. \u003cstrong\u003e83\u003c/strong\u003e(17): p. 51755-51786.\u003c/li\u003e\n\u003cli\u003eDash, S., et al., \u003cem\u003eBig data in healthcare: management, analysis and prospects.\u003c/em\u003e 2019. \u003cstrong\u003e6\u003c/strong\u003e(1): p. 1-25.\u003c/li\u003e\n\u003cli\u003eLv, Z. and L.J.F.G.C.S. Qiao, \u003cem\u003eAnalysis of healthcare big data.\u003c/em\u003e 2020. \u003cstrong\u003e109\u003c/strong\u003e: p. 103-110.\u003c/li\u003e\n\u003cli\u003eBao, S., et al. \u003cem\u003eTechnology enablers for big data, multi-stage analysis in medical image processing\u003c/em\u003e. In \u003cem\u003e2018, the IEEE International Conference on Big Data (Big Data)\u003c/em\u003e. 2018. IEEE.\u003c/li\u003e\n\u003cli\u003eUllah, R. and T.J.A.S. Arslan, \u003cem\u003ePySpark-based optimization of microwave image reconstruction algorithm for head imaging big data on high-performance computing and Google Cloud Platform.\u003c/em\u003e 2020. \u003cstrong\u003e10\u003c/strong\u003e(10): p. 3382.\u003c/li\u003e\n\u003cli\u003eFan, W., D. Zhao, and S. Wang. \u003cem\u003eA fast statistics and analysis solution for medical service big data\u003c/em\u003e. In \u003cem\u003e2015, the 7th International Conference on Information Technology in Medicine and Education (ITME)\u003c/em\u003e. 2015. IEEE.\u003c/li\u003e\n\u003cli\u003eKekevi, U. and A.A.J.C.S. Aydın, \u003cem\u003eReal-time big data processing and analytics: Concepts, technologies, and domains.\u003c/em\u003e 2022. \u003cstrong\u003e7\u003c/strong\u003e(2): p. 111-123.\u003c/li\u003e\n\u003cli\u003eMaxim, B., K.J.I.J.o. C.S. Tetiana, and N. Security, \u003cem\u003eHazelcast Vs. Ignite: Opportunities for Java Programmers.\u003c/em\u003e 2022. \u003cstrong\u003e22\u003c/strong\u003e(2): p. 406-412.\u003c/li\u003e\n\u003cli\u003eAli, A.H. and M.Z.J.K.I.J.o. M.S. Abdullah, \u003cem\u003eA parallel grid optimization of SVM hyperparameter for big data classification using Spark Radoop.\u003c/em\u003e 2020. \u003cstrong\u003e6\u003c/strong\u003e(1): p. 3.\u003c/li\u003e\n\u003cli\u003eStan, C.-S., et al. \u003cem\u003eApache Spark and Apache Ignite performance analysis\u003c/em\u003e. In \u003cem\u003e2019, the 22nd international conference on control systems and computer science (CSCS)\u003c/em\u003e. 2019. IEEE.\u003c/li\u003e\n\u003cli\u003eRallapalli, S. and R.J.I.J.o. A.E.R. Gondkar, \u003cem\u003eApache Spark and Hadoop-Based Big Data Processing System for Clinical Research.\u003c/em\u003e 2018. \u003cstrong\u003e13\u003c/strong\u003e(10): p. 7488-7492.\u003c/li\u003e\n\u003cli\u003eGoyal, S.J.A.S. and E.J.F.A. Research, \u003cem\u003eBig Data and Computer-Human Interaction for Real-Time Illness Diagnosis.\u003c/em\u003e 2024. \u003cstrong\u003e3\u003c/strong\u003e(1): p. 20-31.\u003c/li\u003e\n\u003cli\u003eAl-Quraishi, T., et al., \u003cem\u003eBig Data Predictive Analytics for Personalized Medicine: Perspectives and Challenges.\u003c/em\u003e 2024. \u003cstrong\u003e2024\u003c/strong\u003e: p. 32-38.\u003c/li\u003e\n\u003cli\u003eSmith, J.T., et al., \u003cem\u003eDeep learning in macroscopic diffuse optical imaging.\u003c/em\u003e 2022. \u003cstrong\u003e27\u003c/strong\u003e(2): p. 020901-020901.\u003c/li\u003e\n\u003cli\u003eEbada, A.I., et al., \u003cem\u003eApplying Apache Spark on streaming big data for health status prediction.\u003c/em\u003e 2022. \u003cstrong\u003e70\u003c/strong\u003e(2): p. 3511-3527.\u003c/li\u003e\n\u003cli\u003ePradhan, A.K., et al., \u003cem\u003eOptimizing CNN‐LSTM hybrid classifier using HCA for biomedical image classification.\u003c/em\u003e 2023. \u003cstrong\u003e40\u003c/strong\u003e(5): p. e13235.\u003c/li\u003e\n\u003cli\u003eQuy, N.M., et al., \u003cem\u003eEdge computing for real-time Internet of Things applications: Future internet revolution.\u003c/em\u003e 2023. \u003cstrong\u003e132\u003c/strong\u003e(2): p. 1423-1452.\u003c/li\u003e\n\u003cli\u003eKumar Sharma, N., et al., \u003cem\u003eClassification and Ensemble Machine Learning Techniques to Improve Healthcare Decision Making For Heart Disease.\u003c/em\u003e 2024. \u003cstrong\u003e16\u003c/strong\u003e(1): p. 1-11.\u003c/li\u003e\n\u003cli\u003eBartkov, M., et al., \u003cem\u003eHazelcast Vs. Ignite: Opportunities for Java Programmers.\u003c/em\u003e 2022. \u003cstrong\u003e22\u003c/strong\u003e(2): p. 406-412.\u003c/li\u003e\n\u003cli\u003eAyon, S.I., M.M. Islam, and M.R.J.I.J.o.R. Hossain, \u003cem\u003eCoronary artery heart disease prediction: a comparative study of computational intelligence techniques.\u003c/em\u003e 2022. \u003cstrong\u003e68\u003c/strong\u003e(4): p. 2488-2507.\u003c/li\u003e\n\u003cli\u003eDutta, A., et al., \u003cem\u003eAn efficient convolutional neural network for coronary heart disease prediction.\u003c/em\u003e 2020. \u003cstrong\u003e159\u003c/strong\u003e: p. 113408.\u003c/li\u003e\n\u003cli\u003eMehmood, A., et al., \u003cem\u003ePrediction of heart disease using deep convolutional neural networks.\u003c/em\u003e 2021. \u003cstrong\u003e46\u003c/strong\u003e(4): p. 3409-3422.\u003c/li\u003e\n\u003cli\u003eKiran, J., N. Debbarma, and S. Ganjala. \u003cem\u003eHeart Disease Prediction Using Machine Learning\u003c/em\u003e. In the \u003cem\u003eInternational Conference on Frontiers of Intelligent Computing: Theory and Applications\u003c/em\u003e. 2022. Springer.\u003c/li\u003e\n\u003cli\u003eNancy, A.A., et al., \u003cem\u003eIoT-cloud-based smart healthcare monitoring system for heart disease prediction via deep learning.\u003c/em\u003e 2022. \u003cstrong\u003e11\u003c/strong\u003e(15): p. 2292.\u003c/li\u003e\n\u003cli\u003eGuleria, P., et al., \u003cem\u003eXAI framework for cardiovascular disease prediction using classification techniques.\u003c/em\u003e 2022. \u003cstrong\u003e11\u003c/strong\u003e(24): p. 4086.\u003c/li\u003e\n\u003cli\u003eMienye, I.D. y Y. Sun. \u003cem\u003eEffective feature selection for improved prediction of heart disease\u003c/em\u003e. In the \u003cem\u003ePan-African Artificial Intelligence and Smart Systems Conference\u003c/em\u003e. 2021. Springer.\u003c/li\u003e\n\u003cli\u003eYadav, A.L., K. Soni, and S. Khare. \u003cem\u003eHeart disease prediction using machine learning\u003c/em\u003e. In \u003cem\u003e2023, the 14th International Conference on Computing, Communication and Networking Technologies (ICCCNT)\u003c/em\u003e. 2023. IEEE.\u003c/li\u003e\n\u003cli\u003eHossain, M.M., et al., \u003cem\u003eCardiovascular disease identification using a hybrid CNN-LSTM model with explainable AI.\u003c/em\u003e 2023. \u003cstrong\u003e42\u003c/strong\u003e: p. 101370.\u003c/li\u003e\n\u003cli\u003eBizimana, P.C., et al., \u003cem\u003eAutomated heart disease prediction using improved explainable learning-based technique.\u003c/em\u003e 2024: p. 1-30.\u003c/li\u003e\n\u003cli\u003eMienye, I.D. and N.J.I. Jere, \u003cem\u003eOptimized ensemble learning approach with explainable AI for improved heart disease prediction.\u003c/em\u003e 2024. \u003cstrong\u003e15\u003c/strong\u003e(7): p. 394.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Hadoop, real-time processing, medical big data, Internet of Medical Things, deep learning","lastPublishedDoi":"10.21203/rs.3.rs-7564813/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7564813/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eToday, with the increasing development of technologies, the use of the Internet of Medical Things (IOMT) has created a new revolution in healthcare systems; therefore, it is essential to implement an efficient, scalable method for real-time processing of medical big data for disease monitoring, diagnosis, prevention, and the overall maintenance of patient health. In this paper, we present a hybrid deep learning model, \"CNN-GRU,\" based on gated recurrent units (GRU) and convolutional neural networks (CNN) for processing IOMT data with a distributed processing environment leveraging Apache Spark and Apache Ignite. The proposed architecture uses Spark for parallel and distributed data preprocessing and feature extraction, and Ignite for in-memory computation and low-latency storage. The proposed method's results are evaluated using four parameters: precision, recall, accuracy, and F1 score. Within the suggested model, the average accuracy and recall criterion is 99.01%, the precision criterion is 99.06%, and the F1 score is 99%\u0026mdash;the execution time of the model in three different environments. The Spark, Ignite, and Ignite combinations are 18.84 seconds, 18.93 seconds, and 18.18 seconds, respectively. These values obtained from the results of the hybrid architecture indicate an increase in the efficiency of the proposed method compared to previous approaches. This hybrid architecture offers a clear perspective for personalized healthcare systems.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e","manuscriptTitle":"A Hybrid CNN–GRU Framework for Real-Time IOMT Data Processing Using Distributed Apache Spark and Apache Ignite","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-25 11:15:58","doi":"10.21203/rs.3.rs-7564813/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-10-29T06:27:29+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-15T17:18:16+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-03T10:30:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"90593906653777755894839980574781708220","date":"2025-10-02T13:10:27+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-18T11:56:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"269129096287102941352987815494908766996","date":"2025-09-18T11:40:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"92888702412067634717444977453654184939","date":"2025-09-16T18:43:48+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-16T10:53:47+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-16T10:41:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-10T01:19:53+00:00","index":"","fulltext":""},{"type":"submitted","content":"Cluster Computing","date":"2025-09-08T13:30:06+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"dde5993b-a569-4352-9a28-48d168d8b136","owner":[],"postedDate":"September 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-23T14:08:31+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-25 11:15:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7564813","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7564813","identity":"rs-7564813","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}